BYU Law Review

Volume 2009 Issue 4 Article 4

11-1-2009

The Knowledge Problem of New Paternalism

Mario J. Rizzo

Douglas Glen Whitman

Follow this and additional works at: https://digitalcommons.law.byu.edu/lawreview

Part of the Behavioral Economics Commons, Economic Policy Commons, and the Law Commons

Recommended Citation
Mario J. Rizzo and Douglas Glen Whitman, The Knowledge Problem of New Paternalism, 2009 BYU L. Rev.
905 (2009).
Available at: https://digitalcommons.law.byu.edu/lawreview/vol2009/iss4/4

This Article is brought to you for free and open access by the Brigham Young University Law Review at BYU Law
Digital Commons. It has been accepted for inclusion in BYU Law Review by an authorized editor of BYU Law Digital
Commons. For more information, please contact [email protected].
DO NOT DELETE 11/3/2009 10:03 AM

The Knowledge Problem of New Paternalism

Mario J. Rizzo and Douglas Glen Whitman

It is a standing topic of complaint, that a man knows too little

of himself. Be it so: but is it so certain that the legislator must
know more? It is plain, that of individuals the legislator can
know nothing: concerning those points of conduct which depend
upon the particular circumstances of each individual, it is
plain, therefore, that he can determine nothing to advantage.
Jeremy Bentham1
It may be admitted that, so far as scientific knowledge is
concerned, a body of suitably chosen experts may be in the best
position to command all the best knowledge available . . . .
[Yet] scientific knowledge is not the sum of all knowledge. . . .
[A] little reflection will show that there is . . . the knowledge of
the particular circumstances of time and place. It is with
respect to this that practically every individual has some
advantage over all others in that he possesses unique
information of which beneficial use might be made, but of
which use can be made only if the decisions depending on it are
left to him or are made with his active cooperation.
Friedrich A. Hayek2

I. INTRODUCTION ...................................................................... 907

II. PATERNALIST WELFARE STANDARDS ..................................... 911
III. PATERNALIST POLICIES ALLEGEDLY JUSTIFIED BY
BEHAVIORAL ECONOMICS................................................ 912
A. Sin Taxes ...................................................................... 912
B. Default Enrollment in Savings Plans ............................. 914
C. Cooling-Off Periods ..................................................... 915
D. Risk Narratives ............................................................. 916

 Department of Economics, New York University.

 Department of Economics, California State University, Northridge.
1. JEREMY BENTHAM, AN INTRODUCTION TO THE PRINCIPLES OF MORALS AND
LEGISLATION 319 (W. Harrison ed., Hafner Press 1948) (1781).
2. F.A. Hayek, The Use of Knowledge in Society, 35 AM. ECON. REV. 519, 521–22
(1945).

905
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

E. Employee-Friendly Terms in Labor Contracts .............. 917

IV. A BRIEF THEORY OF PREFERENCE, CHOICE, AND WELFARE . 919
V. THE PATERNALIST’S DILEMMA .............................................. 921
VI. IDENTIFYING THE AGENT’S TRUE PREFERENCES .................. 922
A. Local Knowledge of True Preferences ........................... 922
B. Conflicting Preference Sets ........................................... 924
1. Hyperbolic discounting .......................................... 924
2. Framing and endowment effects ............................. 928
3. Hot and cold states ................................................ 929
VII. DISCOVERING THE EXTENT OF DECISION-MAKING BIAS .... 932
A. Lack of Precision in Measuring Extent of Bias .............. 932
1. Hyperbolic discounting. ......................................... 932
2. Status quo bias ....................................................... 935
3. Endowment effects ................................................. 937
4. Hot and cold states ................................................ 938
5. Optimism and availability bias ................................ 940
B. Absence of a Single Measure of Bias, Even
Intrapersonally ........................................................... 941
1. Hyperbolic discounting .......................................... 941
2. Hot and cold states ................................................ 942
VIII. ACCOUNTING FOR SELF-DEBIASING .................................. 943
A. The Many Varieties of Self-Debiasing and Self-
Regulation ................................................................. 943
1. Cognitive strategies ................................................ 944
2. Environmental strategies ........................................ 945
3. Directly behavioral strategies .................................. 946
B. The Significance of Context for Self-Regulation ........... 946
1. Self-regulation in the laboratory versus in the wild .. 947
2. The automaticity of unconscious self-regulation ..... 949
IX. ACCOUNTING FOR INTERDEPENDENT BIASES ...................... 951
A. Qualitative Effects ........................................................ 952
B. Quantitative Effects ...................................................... 953
X. ANTICIPATING UNRAVELING OF SELF-REGULATION AND
THE SPREAD OF BIASES ..................................................... 955
A. Substitution Effects Between Internal and External
Debiasing ................................................................... 955
B. Generalized Reduction of Self-Regulation .................... 957
XI. ACCOUNTING FOR HETEROGENEITY: THE ONE-SIZE-FITS-
ALL PROBLEM .................................................................. 960
A. Problems of Over-Inclusion and Under-Inclusion ........ 960

906
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

B. Heterogeneity on Multiple Dimensions ........................ 964

1. Fraction of individuals exhibiting a type of bias....... 964
2. Extent of bias ......................................................... 964
3. Extent of self-debiasing .......................................... 965
4. Degree of responsiveness to corrective measures ..... 965
5. Susceptibility of self-debiasing to unraveling ........... 965
XII. CONCLUSIONS: THE ROAD BACK TO OLD PATERNALISM ... 965

I. INTRODUCTION
Recent work in behavioral economics has given rise to a new
theoretical basis for paternalist government policies.3 The literature
of behavioral economics claims that individuals may not always act in
their own best interests. People are not fully “rational,” as
economists understand that term, because their choices are adversely
affected by various cognitive biases, insufficient willpower, and
difficulties of information processing. To the extent that such
decision-making problems are systematic, the claim is made that
deliberate structuring of decision contexts—such as by assigning
appropriate default options, providing cooling-off periods for
commitments, imposing sin taxes, and so forth—can in principle
enhance individuals’ welfare.
The “new” paternalism purports to differ significantly from more
traditional paternalism. The “old” paternalism, which often grew out
of moral or religious notions of the good, effectively ignored the
preferences (or interests or pleasures) of the individual in favor of the
preferences of the policymaker. It does not matter if the individual
really enjoys consuming alcohol, says the old paternalism, because

3. See generally Colin Camerer, Samuel Issacharoff, George Loewenstein, Ted

O’Donoghue & Mathew Rabin, Regulation for Conservatives: Behavioral Economics and the
Case for “Asymmetric Paternalism,” 151 U. PA. L. REV. 1211 (2003); Jonathan Gruber &
Botond Köszegi, Is Addiction ‘Rational’? Theory and Evidence, 116 Q. J. ECON. 1261 (2001);
Christine Jolls & Cass R. Sunstein, Debiasing Through Law, 35 J. LEGAL STUD. 199 (2006);
Ted O’Donoghue & Matthew Rabin, Optimal Sin Taxes, 90 J. PUB. ECON. 1825 (2006)
[hereinafter O’Donoghue & Rabin, Optimal Sin Taxes]; Ted O’Donoghue & Matthew Rabin,
Studying Optimal Paternalism, Illustrated by a Model of Sin Taxes, 93 AM. ECON. REV. 186
(2003) [hereinafter O’Donoghue & Rabin, Studying Optimal Paternalism]; Cass R. Sunstein
& Richard H. Thaler, Libertarian Paternalism Is Not an Oxymoron, 70 U. CHI. L. REV. 1159
(2003) [hereinafter Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron]; Richard
H. Thaler & Cass R. Sunstein, Libertarian Paternalism, 93 AM. ECON. REV. 175 (2003)
[hereinafter Thaler & Sunstein, Libertarian Paternalism].

907
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

that is simply a bad preference. The new paternalism, by contrast,

takes the individual’s own subjective preferences as the basis for
policy recommendations. New paternalist policies allegedly help the
individual to better achieve his own subjective well-being, which
cognitive impediments prevent him from attaining on his own. The
individual who drinks to excess, for instance, may actually be
harming himself by his own internal standards, which he needs help
in meeting because of his lack of willpower.
Many have raised objections to the use of behavioral economics
to justify paternalism. At the most abstract level, there are serious
philosophical questions about the welfare standards implicit in the
new paternalism.4 For instance, is the appropriate goal to maximize
the hedonic satisfaction of agents, or to maximize the satisfaction of
subjective preferences that can transcend hedonic considerations?5 In
this Article, however, we wish to set aside the philosophical
critique—at least as much as possible—and focus on a question of
application: can policymakers reasonably be expected to implement
welfare-improving paternalist policies?
Even this question is too broad, because paternalist policymaking
can be criticized in various ways. A public-choice critique of the new
paternalism would ask whether policymakers have the right
incentives to implement wise policies, given their own self-interest
and the lobbying efforts of interested parties.6 A comparative
institutional critique would observe that policymakers also have
cognitive biases that could inhibit good decision-making.7 A dynamic
critique would highlight the potential for a slippery slope from

4. See Gregory Mitchell, Libertarian Paternalism Is an Oxymoron, 99 NW. U. L. REV.

1245, 1260–69 (2005).
5. See generally Mario J. Rizzo & Douglas Glen Whitman, Meet the New Boss, Same
as the Old Boss: A Critique of the New Paternalism (Jan. 3, 2007) (unpublished manuscript,
on file with New York Univ.).
6. See Edward L. Glaeser, Paternalism and Psychology, 73 U. CHI. L. REV. 133, 144–49
(2006).
7. See generally id. For more informal critiques of the new paternalism (especially
Sunstein & Thaler’s Libertarian Paternalism), see the weblog posts: Posting of Gary Becker,
Libertarian Paternalism: A Critique—BECKER, The Becker-Posner Blog,
http://www.becker-posner-blog.com/archives/2007/01/libertarian_pat_1.html (Jan. 14,
2007, 22:07 CST); Posting of Richard Posner, Libertarian Paternalism—Posner’s Comment,
The Becker-Posner Blog, http://www.becker-posner-blog.com/archives/2007/01/
libertarian_pat.html (Jan. 14, 2007, 21:58 CST).

908
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

modest paternalist policies to more intrusive ones.8 Again, we will set

all these critiques aside, because our question is narrower: do
policymakers have access to the knowledge needed to implement welfare-
improving paternalist policies? The answer, we argue, is no.
The title of this Article is inspired by the “knowledge problem”
identified by Friedrich A. Hayek in his critique of socialist central
planning.9 In the early twentieth century, the advocates of socialism
argued that, in principle, a central planner—equipped with all
relevant knowledge of resource endowments, technologies, and
preferences—could design an efficient economic plan for society.10 In
response, Hayek said that to assume the central planner possesses all
the relevant information about endowments, technologies, and
preferences is to assume the problem away.11 The critical problem that
any economic system must solve is to mobilize and use knowledge
that “never exists in concentrated or integrated form, but solely as
the dispersed bits of incomplete and frequently contradictory
knowledge which all the separate individuals possess.”12 A signal
virtue of a voluntary market order is that it creates conditions under
which such information is more likely to be mobilized and used.
The most important reason that many economists had failed to
appreciate this knowledge problem is that they had been deceived by
their own excessively simple models. They had taken models useful
in understanding some limited features of the real world, such as the
equilibrium reaction of markets to supply or demand shocks, and
applied them to the broader problem of substituting government
planning for market processes. They were guilty of the fallacy of the

8. See Douglas Glen Whitman & Mario J. Rizzo, Paternalist Slopes, 2 N.Y.U. J.L. &
LIBERTY 411, 412–13 (2007); Mario J. Rizzo & Douglas Glen Whitman, Little Brother is
Watching You: New Paternalism on the Slippery Slopes, 51 ARIZ. L. REV. (forthcoming 2009).
9. F. A. Hayek, Socialist Calculation I: The Nature and History of the Problem (1935),
in COLLECTIVIST ECONOMIC PLANNING (F.A. Hayek ed., 1935), reprinted in F. A. HAYEK,
INDIVIDUALISM AND ECONOMIC ORDER 119 (1948); F. A. Hayek, Socialist Calculation II:
The State of the Debate (1935), reprinted in HAYEK, INDIVIDUALISM AND ECONOMIC ORDER,
supra, at 148; F. A. Hayek, Socialist Calculation III: The Competitive “Solution” (1940),
reprinted in HAYEK, INDIVIDUALISM AND ECONOMIC ORDER, supra, at 181.
10. See Oskar Lange, On the Economic Theory of Socialism: Part One, 4 REV. ECON.
STUD. 53, 68–71 (1936); Oskar Lange, On the Economic Theory of Socialism: Part Two, 4 REV.
ECON. STUD. 123 (1937); A.P. Lerner, A Note on Socialist Economics, 4 REV. ECON. STUD. 72
(1936). See generally Hayek, The Use of Knowledge in Society, supra note 2.
11. See Hayek, The Use of Knowledge in Society, supra note 2, at 519.
12. Id.

909
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

misplaced concrete: simple models were mistaken for a simple

world.13
We argue that the new paternalism spawned by behavioral
economics faces a very similar knowledge problem and for similar
reasons. If well-meaning policymakers possess all the relevant
information about individuals’ true preferences, their cognitive
biases, and the choice contexts in which they manifest themselves,
then policymakers could potentially implement paternalist policies
that improve the welfare of individuals by their own standards. But
lacking such information, we cannot conclude that actual paternalism
will make their decisions better; under a wide range of circumstances,
it will even make them worse. New paternalists have not taken the
knowledge problems that are evident from the underlying behavioral
and economic research seriously enough.
To begin, we focus the discussion by outlining several specific
policies that authors in the new paternalist literature have advocated,
as well as the welfare standards and cognitive biases that allegedly
justify them. These policies are chosen as illustrative of new
paternalist policies more generally. The remainder of the Article
uncovers a series of knowledge-based obstacles that paternalist
policies must overcome in order to be effective and justified.
Specifically, paternalist policymakers must (1) identify agents’
“true” preferences that are to be maximally satisfied; (2) determine
the extent of each cognitive bias or decision-making problem; (3)
properly account for privately adopted self-debiasing measures, as
well as how paternalist policies would affect such measures; (4) deal
with the problem of interdependent biases; (5) anticipate unraveling
and unlearning effects; and (6) account for heterogeneity in the
population with respect to all of these factors. We argue that these
factors taken together present a formidable barrier that robs the new
paternalism of any presumption of welfare improvement—even if the
underlying theory and empirical results of behavioral economics are
granted. Furthermore, paternalist policymakers who lack the
information needed to implement policies that actually assist
individuals according to their own subjective preferences will tend to
substitute their own.

13. See generally F. A. HAYEK, THE COUNTER-REVOLUTION OF SCIENCE (1952).

910
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

II. PATERNALIST WELFARE STANDARDS

The “new paternalist” literature, as we shall call it, emphasizes
the possibility of making individuals “better off” according to their
own preferences. Richard Thaler and Cass Sunstein, for instance,
adopt a welfare standard defined in terms of what people would do if
they were perfectly rational:
We intend “better off” to be measured as objectively as possible,
and we clearly do not always equate revealed preference with
welfare. That is, we emphasize the possibility that in some cases
individuals make inferior choices, choices that they would change if
they had complete information, unlimited cognitive abilities, and no
lack of willpower.14
Similarly, Colin Camerer and coauthors choose as their welfare
standard the decisions that individuals would make if they were fully
rational, defined as follows:
First, people have well-defined preferences (or goals) and make
decisions to maximize those preferences. Second, those preferences
accurately reflect (to the best of the person’s knowledge) the true
costs and benefits of the available options.15 Third, in situations
that involve uncertainty, people have well-formed beliefs about
how uncertainty will resolve itself, and when new information
becomes available, they update their beliefs using Bayes’s law—the
presumed ability to update probabilistic assessments in light of new
information.16
The essential problem, as the new paternalists see it, is that
individuals are unlikely to pursue choices that are “in their best
interest”17 in many cases because of cognitive or behavioral biases.
These include “self-control problems,” “fail[ure] to process

14. Thaler & Sunstein, Libertarian Paternalism, supra note 3, at 175 (emphasis added).
15. The second criterion seems to suggest that the agent can have less than complete
knowledge so long as he makes efficient use of his incomplete knowledge. This means that true
preferences are simply optimally-informed preferences. Therefore, for true preferences, in this
attenuated sense, to be different from actual preferences requires that the real-world individual
have less than socially-optimal incentives to acquire information. The authors do not expand
on this point. To use this criterion as a standard for policy intervention would require the
preference paternalist to stop short of complete information in determining true preferences.
How far short would be difficult to assess both theoretically and empirically.
16. Camerer et al., supra note 3, at 1214–15 (2003).
17. Thaler & Sunstein, Libertarian Paternalism, supra note 3, at 175; see also Camerer
et al., supra note 3, at 1212.

911
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

information as Bayes’s rule would require,” and “systematic

mispredictions about the costs and benefits of choices.”18

III. PATERNALIST POLICIES ALLEGEDLY JUSTIFIED BY BEHAVIORAL

ECONOMICS
A wide variety of paternalistic policies could potentially be
justified using these welfare standards, but we cannot address all of
them. We will therefore rely, when necessary, on five illustrative
proposals: sin taxes, default enrollment in savings plans, cooling-off
periods for consumer purchases, risk narratives to accompany risky
products, and employee-friendly terms in labor contracts. With each
proposal, we also discuss the decision-making problems identified by
behavioral economics that are used to justify the policy proposals.
The problems with these diverse policies will not be identical, but all
proposals will encounter at least some of the impediments outlined
in the remainder of this Article.

A. Sin Taxes
Some analysts, notably O’Donoghue and Rabin19 and Gruber
and Köszegi,20 propose to impose sin taxes—e.g., a tax on fatty
foods—to induce better behavior.
The behavioral justification for these sin taxes is that individuals
are afflicted by present-bias or insufficient willpower. Very simply
put, individuals place too much weight on the present relative to the
future.21 This creates a bias toward getting benefits now and
incurring costs later: people spend too much and save too little, they
consume too much and exercise too little, they procrastinate, they
become addicted to drugs, and so on.22

18. Camerer et al., supra note 3, at 1217–18.

19. See O’Donoghue & Rabin, Optimal Sin Taxes, supra note 3; O’Donoghue & Rabin,
Studying Optimal Paternalism, supra note 3.
20. Gruber & Köszegi, supra note 3.
21. For some reason the problem of placing too much weight on the future relative to
the present (hyperopia) is ignored. See, e.g., Ran Kivetz & Anat Keinan, Repenting Hyperopia:
An Analysis of Self-Control Regrets, 33 J. CONSUMER RES. 273, 282 (2006) (concluding that
“consumers sometimes suffer from excessive farsightedness” and “repent hyperopia in the long
run”).
22. Shane Frederick, George Loewenstein & Ted O’Donoghue, Time Discounting and
Time Preference: A Critical Review, 40 J. ECON. LITERATURE 351, 393–94 (2002).

912
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

This is, however, a simplified account. In traditional economic

theory, there is nothing per se irrational about placing more weight
on the present than the future. Indeed, economic models of
intertemporal choice almost universally assume the individual has
some discount factor (often symbolized with the Greek letter δ) that
he applies to future costs and benefits. For instance, someone with a
discount factor of δ = 0.90 would consider $100 of benefits to be
received in a year to be equivalent to $90 received immediately. The
individual’s discount factor is generally considered a matter of
subjective preference.23
For the individual’s behavior to be internally consistent,
however, the discount factor must be constant. That is, the trade-off
between benefits at time 1 and at time 2 should depend only on
their distance from each other, not on their distance from the
present. Thus, for a person with a discount factor of 0.90, $100 to
be received in two years should be equivalent to $90 in one year,
and $100 to be received in one year should be equivalent to $90
now. This is known as exponential discounting.24 Behavioral research,
however, indicates that real people are inconsistent discounters. For
instance, an individual might regard $100 to be received in two years
as equivalent to $90 to be received in one, and yet he might regard
$100 to be received in a year as worth only $80 now. This
phenomenon is known as hyperbolic discounting.25
People who engage in hyperbolic discounting may exhibit time
inconsistency: they will make decisions about future trade-offs and
then reverse those decisions later. For instance, if offered a choice
between $100 in two years and $85 in one year, the person
described above chooses the larger sum. Yet when a year has passed,
he reverses his prior choice and takes the smaller sum ($85), because
the $100 to be received in a year is regarded as worth only $80 now.

23. Economists sometimes refer to a discount rate instead of a discount factor. The
discount rate is related to the discount factor in the following way: r = (1 – δ)/δ. Throughout
this Article, we will use only discount factors.
24. “Exponential” refers to the fact that the discount factor must be multiplied by itself
multiple times to discount events multiple periods in the future. For instance, in the example
given, $100 to be received in two years would be valued at (0.90)(0.90)($100) =
2
(0.90) ($100) = $81 now.
25. See generally Frederick, Loewenstein & O’Donoghue, supra note 22 (reviewing the
relevant literature on experiments of this nature). The seminal article in this literature is R.H.
Strotz, Myopia and Inconsistency in Dynamic Utility Maximization, 23 REV. ECON. STUD. 165
(1955–1956); see also GEORGE AINSLIE, BREAKDOWN OF WILL (2001).

913
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

Behavioral economists take this sort of inconsistency as evidence of

irrationality.26
Thus, proponents of sin taxes use hyperbolic discounting to
explain self-control problems. Intuitively, people’s inconsistent
behavior reflects their vulnerability to temptation when those
temptations are near. With regard to eating, for example, a
hyperbolic discounter might promise to start a diet tomorrow, but
then reverse that decision once tomorrow has become today. A
properly-calibrated sin tax, its new paternalist supporters argue,
would make the overeater fully account for the future costs of her
current choices by increasing the present cost. This increase in
present cost, new paternalists argue, offsets the hyperbolic discount
and aligns the person’s decision with “rationality.”

B. Default Enrollment in Savings Plans

Various authors, but most notably Thaler and Sunstein, have
advocated automatically enrolling new employees in savings plans
from which they could voluntarily opt out (as opposed to the more
common practice of not enrolling employees until they opt in).27 It is
not always clear in the literature whether this recommendation is
directed solely at employers, or if the new paternalists would also
support a government requirement that employers implement
automatic enrollment. Sunstein and Thaler say the law “might
require employers to provide automatic enrollment and allow
employees to opt out.” Further, they say this would be consistent
with their notion of “libertarian paternalism,” but they do not
explicitly advocate this policy.28 Camerer et al. strongly imply that
mandatory savings default rules may be necessary because firms lack
sufficient incentive to offer optimal defaults.29 For this Article, we
will consider a legal mandate on employers to adopt default
enrollment.

26. The dollar values are used here only for illustrative purposes. In principle, the costs
and benefits need not be monetary; they can be pleasures, pains, health effects, and so on. The
key question is how benefits and costs of whatever form are weighed against each other when
they occur at different points in time.
27. See Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra note 3,
at 1159–1202; Thaler & Sunstein, Libertarian Paternalism, supra note 3, at 175–79.
28. Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra note 3, at
1176–77.
29. Camerer et al., supra note 3, at 1251–52.

914
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

One behavioral argument in favor of default enrollment is the

same as that used for sin taxes: individuals are afflicted by hyperbolic
discounting, which causes them to weigh the present too heavily.
The present benefits of greater consumption, combined with the
present costs of going through the enrollment process, induce
individuals to delay enrollment and thus save too little.
More often, however, the case for default enrollment is based on
inertia or status quo bias: the psychological tendency of people to
maintain current arrangements, whatever they might be.30 Sunstein
and Thaler say that employees often fail to enroll under an opt-in
system, but they would choose to enroll if they simply took the time
to think carefully.31 The idea, then, is to place employees into a new
status quo that is more likely to match their considered preferences.32

C. Cooling-Off Periods
There are two types of cooling-off periods. One kind creates a
mandatory waiting period before a purchase or other decision can be
made.33 The other creates a mandatory period following a purchase
or other decision during which it can be reversed by one of the
parties.34 For example, a cooling-off period for marriage requires a
certain number of days to pass between issuance of a marriage license
and the marriage itself; a cooling-off period for new cars allows a car
buyer to return the car within a few days of the sale without
penalty.35

30. For a detailed discussion see William Samuelson & Richard Zeckhauser, Status Quo
Bias in Decision Making, 1 J. RISK & UNCERTAINTY 7, 33–41 (1988).
31. “If employers think (correctly, we believe) that most employees would prefer to join
the 401(k) plan if they took the time to think about it and did not lose the enrollment form,
then by choosing automatic enrollment, they are acting paternalistically by our definition of
the term.” Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra note 3, at
1172–73.
32. Status quo bias and hyperbolic discounting are not always clearly distinguished in
the case for default savings enrollment. For instance, Sunstein and Thaler state that “[e]ven a
trivial action, such as filling in some form and returning it, can leave room for failures due to
memory lapses, sloth, and procrastination.” Id. at 1181. Although they do not specifically
invoke the notion of hyperbolic discounting, that is the leading explanation among behavioral
economists for procrastination in areas such as dieting and saving.
33. Camerer et al., supra note 3, at 1240.
34. Id.
35. Without an ex post penalty, that is. The initial purchase price might be higher to
account for costs associated with having a cooling-off period.

915
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

The behavioral support for cooling-off periods is the evidence

that people make different decisions depending on whether they are
in a “hot” or “cool” state.36 According to Sunstein and Thaler,
“[t]he essential rationale [for cooling-off periods] is that under the
heat of the moment, consumers might make ill-considered or
improvident decisions.”37 Camerer et al. note that this rationale is
supported by evidence that people make costly or even irreversible
choices when they are in a biologically “hot” state (such as anger,
fear, excitement, or sexual arousal) that they would not make if they
were in a “cool state” (calm, reflective, and sober).38 New
paternalists support the cooling-off period because it either forces
the decision maker to delay his decision until he is in a cooler mental
state, or allows him to reconsider once he is in such a state, thus
allowing him to pursue his “true” preferences.

D. Risk Narratives
New paternalists also support the use of “risk narratives” to aid
individuals in making risky decisions. When consumers consider
purchasing dangerous products or engaging in dangerous activities,
they could be informed about the relevant risks by means of
statistical summaries of the likelihood of various harms. Alternatively,
they could be informed by means of accounts, or narratives, about
specific people who have suffered harm from the product or activity
in question. Sunstein and Jolls propose that providers be required by
law to provide such narratives:
Specifically, the law could require firms—on pain of administrative
penalties or tort liability—to provide a truthful account of
consequences that resulted from a particular harm-producing use of
the product, rather than simply providing a generalized warning or
statement . . . .39
We will refer to this policy as “risk narratives.” The behavioral
justification for this policy is that people are afflicted by optimism
bias, which causes them to underestimate their personal likelihood of

36. See George Loewenstein, Emotions in Economic Theory and Economic Behavior, 90
AM. ECON. REV. 426 (2000) [hereinafter Loewenstein, Emotions in Economic Theory].
37. Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra note 3, at
1188.
38. Camerer et al., supra note 3, at 1238–40.
39. Jolls & Sunstein, supra note 3, at 212.

916
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

suffering adverse consequences.40 To take just one example, they

underestimate their chances of getting into an automobile accident.41
As a result, they will be too likely to expose themselves to risks.
Interestingly, Jolls and Sunstein propose to correct optimism bias
by exploiting a different bias: the availability heuristic,42 which refers
to the tendency to judge the probability of an event based on “an
assessment of how easily examples of the event can be called to
mind.”43 For instance, a person whose grandmother died from a rare,
but not genetic, illness might overestimate the likelihood of
contracting that illness—simply because it happened to someone he
knows. Thus, risk narratives harness the availability heuristic to
counter optimism bias—a person’s overestimation of risk upon
hearing a general warning is offset by an underestimation of risk
caused by exposure to vivid stories about harmed individuals. The
result, claim the new paternalists, should be an accurate risk
calculation and an appropriate decision.

E. Employee-Friendly Terms in Labor Contracts

Sunstein and Thaler suggest various terms that could be included
in labor contracts for the benefit of employees. For instance, they
suggest making “for cause” rather than “at will” the default
termination rule,44 lengthening the presumed amount of paid
vacation time,45 and presuming protection against age discrimination
unless the employee waives such protection.46
In addition to these suggestions, in which the defaults are fully
waivable, Sunstein and Thaler suggest other policies (including some
existing policies) that are only partially waivable. For instance, they
support the provision of the Model Employment Termination Act,

40. W. KIP VISCUSI & WESLEY A. MAGAT, LEARNING ABOUT RISK: CONSUMER AND
WORKER RESPONSE TO HAZARD INFORMATION 95–96 (1987); Christine Jolls, Behavioral
Economic Analysis of Redistributive Legal Rules, 51 VAND. L. REV. 1653, 1659–62 (1998).
41. Jolls & Sunstein, supra note 3, at 205.
42. See generally Amos Tversky & Daniel Kahneman, Judgment Under Uncertainty:
Heuristics and Biases, 185 SCI. 1124 (1974); Amos Tversky & Daniel Kahneman, Availability:
A Heuristic for Judging Frequency and Probability, 5 COGNITIVE PSYCHOL. 207–32 (1973).
43. Jolls & Sunstein, supra note 3, at 204.
44. Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra note 3, at
1175.
45. Id. at 1176.
46. Id. at 1177.

917
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

which replaces “at will” with “for cause” termination.47 This right
can be waived by agreement—but only if the employer agrees to
provide a severance payment (which the Model Act sets as “one
month’s salary for every year of employment”) in the event of a not-
for-cause termination.48 Note that the Model Act does not allow
employees to waive their right to “for cause” termination by
negotiating for higher regular salary or for any severance pay less
than one month’s salary per year of employment. Similarly, Sunstein
and Thaler reinforce the case for the Fair Labor Standards Act, which
says that employees may not be required to work beyond 40 hours
per week.49 This provision may be waived in return for time-and-a-
half pay.50 Note that it cannot be waived for any lower rate of pay
(including the regular rate), even if the employer and employee agree
upon it.
The behavioral justification for changing default rules—and
sometimes making the defaults costly to change—is that people are
subject to framing effects.51 This means their decisions tend to be
sensitive to seemingly irrelevant aspects of how the choice situation is
described. Probably the best known type of framing effect is the
endowment effect, which refers to people’s tendency to demand
more compensation to give something up (their willingness to
accept, or “WTA”) than they would have paid to acquire that same
thing (their willingness to pay, or “WTP”).52
For rational agents, the default should not make a difference for
choices (at least if transaction costs are low). But for less rational

47. Id. at 1187. Sunstein and Thaler do qualify their endorsement to some extent by
admitting that provisions with substantive limitations on waiver are “less libertarian than [they]
might be.” Id.
48. Id. at 1187 (citing MODEL EMPLOYMENT TERMINATION ACT §§ 3(a), 4(c),
reprinted in MARK A. ROTHSTEIN & LANCE LIEBMAN, EMPLOYMENT LAW: CASES AND
MATERIALS 211 (Statutory Supp. 2003)).
49. Id.
50. Id. (citing 29 U.S.C. § 207(f) (2000)).
51. See generally Cleotilde Gonzalez et al., The Framing Effect and Risky Decisions:
Examining Cognitive Functions with fMRI, 26 J. ECON. PSYCHOL. 1 (2005), available at
http://repository.cmu.edu/cgi/viewcontent.cgi?article=1012&context=sds.
52. See Jolls & Sunstein, supra note 3, at 205. For instance, students “endowed” with a
university mug demanded more to part with the mug than they would have paid to buy it. Id.
Given the mug’s low value relative to the students’ wealth, the two situations are effectively
identical: they are being asked to choose between a mug and money. Regardless of whether
they were given the mug to begin with, both mug and money were options. Yet the students’
choices differed.

918
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

agents, the default rule can matter. Employees, for example, might
demand more compensation to eliminate a “for cause” termination
clause than they would sacrifice to insert it. They might demand
more compensation to give up additional vacation time than they
would sacrifice to acquire it. The idea, then, is to structure defaults
in labor contracts to increase the likelihood of employees getting
favorable terms.

IV. A BRIEF THEORY OF PREFERENCE, CHOICE, AND WELFARE

In order to convey the underlying unity of our knowledge-based
critique of new paternalist policy prescriptions, it is useful to outline
our views on the evidential character of individual choice. As we have
seen, the core of the behavioral critique of standard welfare theory is
the claim that individuals do not always reveal their true subjective
preferences through their actual choices (because of inadequate
willpower or knowledge or both). We do not deny this claim about
actual choice.
However, the possibility that an individual’s particular choices
can be erroneous in this sense does not mean that we must abandon
actual choice as the ultimate evidence of welfare. It simply means
that we must be more inclusive about which choices are relevant.
Individuals may be aware of their own lack of self-control or that
they make systematic mistakes. When that is the case, we would
expect them to make choices that manifest this awareness. For
instance, individuals may bind themselves ex ante or acquire better
information about the consequences of their actions. This
perspective requires us to observe a complex of choices rather than a
single choice. In order to make sense of the behavior at issue in this
way, we must maintain the hypothesis that some of these related
choices do, in fact, express the related preferences of individuals.53
For example, suppose a worker is paid in cash. On his way home, he
is tempted to stop by a bar and drink away a good part of his salary.
Since he often succumbs to this temptation, he may eventually take
steps to avoid it. He may choose to take a different route home,
bypassing the bar. He may request that his salary not be paid in cash

53. A maintained hypothesis is simply held for the moment. It may be questioned under
different circumstances or at different times. Thus it does not amount to the view that any
particular choice is privileged insofar as it necessarily reveals the subjective preferences of the
individual.

919
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

or that his employer directly deposit his salary in a bank account.

These choices, in conjunction with the choice-at-issue, reveal that he
has a self-control problem.54 But we cannot infer simply from the
choice to drink away a good portion of his salary that his choice
displays a lack of willpower or is otherwise “irrational.”
The choices that reveal he has a problem, however, are
simultaneously the choices that attempt to solve the problem. Does our
approach, therefore, imply that the individual’s solution is always
sufficient? Can someone else know if there is a better way? These are
separate questions. Obviously, the individual’s solution may not be
sufficient. There may be unexploited opportunities for gain; better
techniques of self-control may exist. But how would anyone else
know? For another to know, he must know the benefit to the
individual of controlling his behavior as well as the cost of self-control
mechanisms. Since the other person does not have access to the
contents of the individual’s mind, the only way to know is to offer a
wider array of techniques to the individual and observe if he chooses
one of them. But absent an actual choice by the individual, we
cannot know.
Some have argued that since statements are speech-acts, we can
use what people say about their choices as evidence of the possible
irrationality of those choices.55 For instance, if an individual says he
wants to lose weight, this would constitute evidence of his actual
preferring to do so—and thus of his “irrationality” in continuing to
overeat. While we should not exclude this possibility entirely, it is
important to point out that speech-acts and other choices generally
have very different cost-benefit structures. The incentives to say
something are not the same as the incentives to do the thing spoken
of.
The individual can say, “I want to save more, but I am too weak-
willed.” What does this mean for purposes of economic analysis or
public policy? It is entirely unclear. Is the individual expressing a
preference or a simple desire? A preference reflects the willingness to

54. Just as the worker has various means to deal with his self-control problem,
acquisition of more information, solicitation of expert advice, and attempts to improve one’s
computational skills plausibly reflect the preference to improve the knowledge content of
decisions.
55. See Andrew Caplin, Economic Theory and Psychological Theory: Bridging the Divide,
in THE FOUNDATIONS OF POSITIVE AND NORMATIVE ECONOMICS: A HANDBOOK 336, 359–
60 (Andrew Caplin & Andrew Schotter eds., 2008).

920
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

incur the opportunity cost, whereas a desire is just a generally

favorable attitude toward something irrespective of opportunity cost.
The statement itself does not reveal a serious willingness to incur the
opportunity cost of more savings. It is evidence simply of his
willingness to incur the costs of the statement to attain its benefits.56
The saying and the doing are different actions. Saying is not by itself
evidence of true and comprehensive underlying preferences.

V. THE PATERNALIST’S DILEMMA

To understand the fundamental problem facing the paternalist,
recall that the rationale for new paternalism is that the individual has
cognitive and behavioral limitations that prevent him from either
recognizing difficulties in the pursuit of his welfare or efficiently
overcoming them. This implies a complex interrelationship between
the knowledge needed by the paternalist and the knowledge
possessed, or capable of being acquired, by the individual.
The paternalist must be smarter than the target agents. He must
know their preferences better than they do in order to know just
what their difficulties are and how they may be efficiently overcome.
Yet both the problems and the solutions are contextual. They
depend on local and personal knowledge. Thus, even if the
paternalist has better theoretical knowledge about cognitive and
behavioral biases, it will be of little use unless he has considerable
local knowledge about a specific individual’s preferences, self-control
problems, available options, and so forth. Ultimately, if the goal is
superior action on the part of agents, the superior theoretical
knowledge of the paternalist cannot be directly relevant to the
individual, except by way of advice. The best course of action for the
individual to take will depend on what Hayek called “knowledge of
the particular circumstances of time and place.”57 This includes
knowledge of locally and temporally contingent external facts, facts
about the individual’s personal traits and, more specifically, facts
about particular temptations and strategies to avoid them.
Sometimes these facts are consciously held and utilized, while in
other cases they may be tacitly or unconsciously held and utilized.

56. Individuals may wish to signal to others or to themselves that they are prudent
without being prudent. Thus they are willing to incur the costs of deception for its benefits.
This does not imply that they are willing to incur the costs of actual prudence for its benefits.
57. Hayek, The Use of Knowledge in Society, supra note 2, at 521.

921
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

This knowledge is largely inaccessible by paternalists and yet, without

it, they cannot use their putatively superior theoretical knowledge to
develop welfare-improving policies.
In the remaining sections of this Article, we will discuss in
greater detail the many kinds of knowledge that paternalist
policymakers would need to have in order to improve individual
choices.

VI. IDENTIFYING THE AGENT’S TRUE PREFERENCES

The issue is whether policymakers—including voters, politicians,
judges, and bureaucrats—can generate general, clearly articulable
rules in the form of taxes, subsidies, cooling-off periods, and so
forth, to counteract what would otherwise be the inferior decisions
of agents. One prerequisite for welfare-improving policies is that
policymakers must possess superior knowledge of people’s “true”
preferences—that is, the preferences the new paternalists allegedly
wish to advance.
Evidence suggests that agents may not have “true” preferences at
all.58 This, in itself, presents a problem for the new paternalist
paradigm; we cannot claim to be making people better according to
their preferences if such preferences do not exist. But we will assume,
arguendo, that true preferences do in fact exist. Let us first address
the general question: Does the paternalist know true preferences
better than the agent himself?

A. Local Knowledge of True Preferences

The relevant question is whether policymakers can be expected
to have better knowledge of true preferences than the agents in
question. Since “better” is defined in terms of the individual’s
subjective welfare (as opposed to old-style paternalism), we must
compare the relative ability of individuals to make welfare-enhancing
decisions for themselves with the ability of outsiders to decide on
their behalf.

58. The new paternalists admit this. See, e.g., Sunstein & Thaler, Libertarian
Paternalism Is Not an Oxymoron, supra note 3, at 1164 (“If the arrangement of the
alternatives has a significant effect on the selections the customers make, then their true
‘preferences’ do not formally exist.”). In general, the question of whether preferences formally
exist will arise whenever individuals exhibit preference reversals or make frame-dependent
choices.

922
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

Lacking direct evidence on the relative ability of individuals

versus government to discover preferences, let us consider the
relative ability of individuals versus close friends and family members.
In a recent study, Joel Waldfogel compares the valuations by
consumers of items they purchased themselves to their valuations of
similar items purchased for them as gifts.59 The gift-givers were
friends and extended and immediate family members—in general,
individuals who would likely have some personal, local, and
sometimes tacit knowledge of the recipients’ preferences.60 The
consumer goods in question were familiar to both giver and receiver,
did not involve intertemporal choices, and were not uncertain in the
sense that people did not think they were buying one thing and got
another.61 In other words, these were relatively “simple”
consumption choices. If the consumers were no better or worse at
determining what satisfied their preferences than the gift-givers, we
should expect the consumers’ ex post evaluations of the self-
purchased items versus gifts to have been about equal, on average.
Instead, “consumers’ own purchases generate[d] between 10% and
18% more value, per dollar spent, than items received as gifts.”62
Is it likely that the ignorance of consumers about their own
preferences made them incapable of accurately evaluating the relative
values of self-purchases and gifts? Perhaps they simply reaffirmed
their decisions in the survey that was undertaken by Waldfogel
shortly after the decisions were made. However, most of the
products purchased by either party were of the type that would likely
show their “true value” rather quickly—sweaters, shirts, books, CDs,
jackets, hats, and so forth.63 So it seems reasonable to expect that
consumers’ ignorance about their own preferences was largely
resolved ex post. If this is so, then it makes sense to use consumer
valuations of the relative efficiency of own-purchases and those of
gift-givers. In view of the evidence that “consumers fare better [at
identifying their own preferences] than all types of givers except

59. Joel Waldfogel, Does Consumer Irrationality Trump Consumer Sovereignty?, 87 REV.
ECON. AND STAT. 691 (2005).
60. Id.
61. Id.
62. Id.
63. Most of the purchases were of this sort, that is, short-run experience goods.
However, the value of a few purchases could not be immediately ascertained, such as
electronics, kitchen appliances, or perhaps video games. See id. at 695 tbl.3.

923
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

significant others and possibly grandparents . . . it seems unlikely that

an alternative chooser would do better than friends, siblings, and
parents, all of whom have substantial amounts of information about
the ultimate consumer’s preferences.”64
In short, even friends and family have a difficult time doing any
better than the individual himself in making welfare-enhancing
choices. Yet friends and family are more likely than policymakers to
have the local knowledge necessary to make wise decisions. Thus,
Waldfogel’s study provides at least suggestive evidence of the
difficulty new paternalists will face in crafting wise policies. The basic
problem is that paternalist policymakers need a baseline of “true”
preferences to satisfy, but the knowledge of such preferences is very
hard to access. That individuals sometimes have difficulty
determining their own preferences does not mean outsiders will do
any better; they can also do worse.

B. Conflicting Preference Sets

We now turn to the more technical question of how
policymakers might go about determining what true or informed
preferences are, assuming once again that they do exist. The case for
a decision-making bias is typically based on the existence of an
inconsistency in individual choices, which presumably corresponds to
an internal inconsistency of preferences. But identifying an
inconsistency in someone’s behavioral preferences (meaning those
that actually determine choice) is not the same as identifying
someone’s true preferences. To do that, we would have to know
which of the inconsistent behavioral preferences better represents the
agent’s actual welfare. We will consider three types of bias for which
this problem arises: hyperbolic discounting, framing and endowment
effects, and hot and cold state effects.

1. Hyperbolic discounting
Sometimes individuals make different choices about present
versus future consumption depending on the time at which the
decision is made, even if the two periods being compared do not
change. To take the example given earlier, in the discussion of sin
taxes, an individual today might choose $100 to be received in two

64. Id. at 695.

924
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

years over $85 to be received in one year, and yet reverse that
decision when a year has passed and choose $85 immediately over
$100 the next year (even though nothing else has changed). This
inconsistency in choice is modeled as an underlying inconsistency in
preferences.65 We could assume—as do Gruber and Köszegi66 and
O’Donoghue and Rabin67—that the true preferences are those
represented by the more far-sighted choice, and the question
becomes how to make the near-term choices correspond to the far-
sighted preferences. But what basis is there for this assumption? We
could just as easily designate the more near-sighted preferences as
the correct ones, and then aim to make far-term choices better
correspond to them.
To put it another way, an internally consistent person would
have a single discount factor. In our example, we have an individual
with two discount factors: 0.90 between any two adjacent years in
the future, and 0.80 between the present year and the next year. This
person exhibits time inconsistency by choosing $100 over $85 when
both are in the future, then reversing that decision by choosing $85
over $100 when the $85 is to be received immediately. One way to
make this person internally consistent would be to make him use a
discount factor of 0.90 for all his intertemporal decisions. Thus, he
must choose the $100 later over the $85 earlier every time. But
another way to make this person internally consistent would be to
make him use a discount factor of 0.80 for all intertemporal
decisions, so that he will choose the $85 earlier over the $100 later
every time. As either of these “corrections” would make the agent’s
behavior consistent, we lack a means of saying which discount factor
corresponds to the agent’s “true” preferences, even if we concede
that one of them must.
To make the problem more vexing, the paternalist may not face a
choice between just two discount factors. The paternalist
policymaker might favor the more far-sighted (larger) discount
factor, the more near-sighted (smaller) discount factor, or some
discount factor that lies somewhere in between, reflecting an

65. Inconsistency in preferences, however, need not produce inconsistency in or reversal

of choices. See HOWARD RACHLIN, THE SCIENCE OF SELF-CONTROL 39 (2000). In this case
we would have “myopia,” that is, a large short-run discount factor and a small long-run
discount factor, without preference reversal.
66. Gruber & Köszegi, supra note 3.
67. O’Donoghue & Rabin, Studying Optimal Paternalism, supra note 3.

925
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

intermediate degree of patience. If there is inconsistency between

two different preference sets, there is no reason for the paternalist to
assume the agent’s “true” preferences must be one of those two.
Moreover, the research on time discounting does not reveal a
simple binary choice process, wherein the individual applies one
discount factor when comparing two future periods and another
discount factor when comparing the present period and a future
period. Instead, the discount factor varies continuously depending
on how far away the nearer period is.68 For example, the individual
might apply the discount factor 0.98 when comparing rewards to be
received ten versus eleven years in the future; the discount factor
0.90 when comparing rewards five versus six years in the future; the
discount factor 0.80 when comparing rewards two versus three years
in the future; and the discount factor 0.70 when comparing rewards
now and a year from now. Thus, in economic terminology,
preferences are truly hyperbolic, not just quasi-hyperbolic.69 If finding
the agent’s true preferences means finding a single time-discounting
factor that can be used as the basis for exponential discounting
(which is time consistent), then the existence of true hyperbolic
discounting means the paternalist has infinitely many different
options to choose from—and no objective means of doing so.
There have, however, been some attempts to justify using the
lower or long-term rate. The first is based on the assumption that
individuals have “stable lifetime preferences” and thus any temporary
deviation from them is a mistake.70 This is reinforced by the idea
that, for most of the future periods about which plans are made, a
higher (more patient) discount factor is applied. Unfortunately, the

68. In his early work, Richard Thaler finds three effective annual discount rates ranging
from 345% over a one-month horizon to 120% over a one-year horizon to only 19% over a
ten-year horizon. Richard H. Thaler, Some Empirical Evidence on Dynamic Inconsistency, 8
ECON. LETTERS 201, 201–07 (1981).
69. See AINSLIE, supra note 25, at 28–35. “Quasi-hyperbolic” refers to a discounting
process involving only two discount factors: one that applies between any two future periods
(lower), and an additional discount that applies between the present and any future period
(higher). “Hyperbolic” refers to a continuously declining discount rate as the future periods of
comparison become more distant. See David Laibson, Golden Eggs and Hyperbolic Discounting,
112 Q.J. ECON. 443, 446–51 (1997).
70. B. Douglas Bernheim & Antonio Rangel, Behavioral Public Economics: Welfare and
Policy Analysis with Non-Standard Decision-Makers, in ECONOMIC INSTITUTIONS AND
BEHAVIORAL ECONOMICS (Peter Diamond & Hannu Vartiainen eds., forthcoming)
(manuscript at 11, 26, available at http://www.stanford.edu/~bernheim/Behavioral%20
Public%20Economics%20Final.pdf).

926
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

idea of stable lifetime preferences is merely an assumption.

Furthermore, to draw normative significance from the stylized fact
that a higher discount factor applies to more periods than does the
lower discount factor seems little more than an attempt to derive an
“ought” from an “is.”
Moreover, the “mistake” interpretation founders on the shoals of
truly hyperbolic discounting. When there are only two discount
factors, it is deceptively simple to designate one of them as “correct.”
When there are infinitely many discount factors, the selection is not
nearly so simple. We suspect that advocates of the “mistake”
interpretation have been misled by the quasi-hyperbolic
approximation, which was originally adopted more for its
mathematical tractability (relative to models of true hyperbolic
discounting) than for its accuracy in describing human behavior.71
The presence of more than two discount factors raises the
possibility that, unless the discount factor representing the highest
degree of patience is always regarded as the appropriate standard,
decision-makers can be too future-oriented as well as too impatient.
They may fail to recognize that life is not forever and may not pluck
enough flowers. Specifically, Kivetz and Keinan have shown in a
number of studies that as temporal perspective lengthens, individual
regret over the failure to seize the pleasures of life grows while guilty
regret over indulgence falls, with the former ultimately
predominating.72
A second attempt to justify using a longer-term rate as the
normative rate is based on the idea that a planning rate is considered
more than the acting rate. In other words, the planning rate is the
result of a calm, collected, and thoughtful process while the acting
rate is dominated by transient passions.73 But this is far from the only
plausible explanation of the difference between short and long-term
discount factors. First, it is not unreasonable to believe that the

71. See AINSLIE, supra note 25, at 210 n.29, 214 n.21; George-Marios Angeletos et al.,
The Hyperbolic Consumption Model: Calibration, Simulation and Empirical Evaluation, 15 J.
ECON. PERSP. 47, 50 (2001).
72. This does not imply that the choice of a long-run benefit (“virtue”) over a short-
term indulgence (“vice”) is always the source of predominant regret in the long run, but that it
can be, especially when the optimal decision is not obvious. See Kivetz & Keinan, supra note
21, at 274.
73. Daniel Read, Which Side Are You On? The Ethics of Self-Command, 27 J. ECON.
PSYCHOL. 681, 685 (2006). Read calls the planner the “pre-agent” and the actor “the agent.”
Id.

927
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

opportunity costs of “virtue” become more evident as the moment of

action arrives.74 Far from being a less considered behavior, the
present-oriented action may therefore actually be more considered
and better informed. Second, Paul Glimcher and coauthors have
found that a lower (more impatient) discount factor is applied
whenever one of the outcomes compared is the earliest possible.75
Specifically, agents apply the same discount factor for a choice
between today and one day later as they do for a choice between
sixty days from now and sixty-one days from now, when the latter is
the earliest possible option.76 This suggests that considered choice may
not be at issue, because the same discount factor is applied even
when the earliest possible date is two months away.
Third, the phenomenon of hyperbolic discounting may not be a
strict matter of time-preference at all. Ariel Rubinstein has argued
that differences in time periods seem more similar when agents
contemplate them in the farther future than in the near term.77
Goods delivered in 101 days or 111 days are more similar to each
other than the same two goods delivered in 1 day or 11 days.78
Under these circumstances the delay-attribute more or less drops out
of consideration in the first case but not in the second. Thus the
more patient “long-run” discount factor is the result of a relative
failure to envision or appreciate future time delays.79 This suggests
that the short-term rate or rates might be more considered.

2. Framing and endowment effects

The framing problem is also evidenced by individuals making
different choices for identical choice problems presented in different
ways. Again, the inconsistency in choice allegedly reveals an
underlying inconsistency of preferences. If we set aside that the

74. “The information available to the acting-agent about the local consequences of a
specific choice will often be better than the information available to the pre-agent. When a
dieter changes his mind and has tiramisu after promising not to, it might be because he is
weak-willed, or it might be because he has only now realized how appealing the tiramisu is.”
Id.
75. See Paul William Glimcher, Joseph Kable & Kenway Louie, Neuroeconomic Studies of
Impulsivity: Now or Just as Soon as Possible?, AM. ECON. REV., May 2007, at 142.
76. Id. at 143–45.
77. Ariel Rubinstein, “Economics and Psychology”? The Case of Hyperbolic Discounting, 44
INT’L ECON. REV. 1207 (2003).
78. Id.
79. Id. at 1210.

928
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

different frames might actually matter to the individual’s subjective

well-being and suppose the frame really is irrelevant80 we still have to
ask: what are the true preferences? If the choice under Frame A
corresponds to preference set A, and the choice under Frame B
corresponds to preference set B, either A or B could represent the
agent’s true preferences. And here, too, the paternalist does not
necessarily face a simple binary choice, as A and B might not
represent the only way to frame the problem.
Take the example of vacation time in employment contracts.
Suppose that potential employees feel different, and thus negotiate
differently, when more vacation time is a default part of the contract
than when it is not. Under default A (less vacation time), added
vacation time seems less valuable, so the employee does not strongly
negotiate for it. Under default B (more vacation time), that added
time seems more valuable, so the employee strongly resists its
reduction.81 This is a classic case of willingness-to-pay differing from
willingness-to-accept, and thus evidence of internally inconsistent
preferences. But which default rule corresponds to the agent’s true
preferences, representing his actual trade-off between leisure time
and money? This question is crucial to the policy choice of an
optimal default, since the wage rate will fall to compensate for longer
vacation periods, yet the theory provides the policymaker with no
means of choosing.82

3. Hot and cold states

The existence of a bias based on emotional states is supposedly
revealed by an individual making different choices depending on
whether he is in a “hot” or “cold” state. For instance, a person may
choose to have sex or eat unhealthily when in a hot (aroused or
hungry) state, yet refuse the same opportunity when in a cold (not-

80. Madrian and Shea argue that framing the 401(k) participation decision in such a way
that enrollment is the default is likely to be seen by employees as “implicit advice” from
employers who presumably know better. Brigitte C. Madrian & Dennis F. Shea, The Power of
Suggestion: Inertia in 401(k) Participation and Savings Behavior, 116 Q.J. ECON. 1149, 1182
(2001). To the extent that this is the case, framing really does matter because it conveys
information. Id.
81. See, e.g., Sunstein & Thaler, Libertarian Paternalism Is Not an Oxymoron, supra
note 3, at 1176.
82. In cases like this, the tendency of the policymaker is to adopt an objective standard
of welfare and set the default to the option that is “objectively” better. This constitutes an
abandonment of the new paternalist project.

929
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

aroused or not-hungry) state. This seems to reveal an

inconsistency—although this is likely a case when the emotional state
itself has a large effect on the actual satisfaction gained from the
activity. But let us suppose a real inconsistency is revealed here: the
sex or junk food would be as physically satisfying either way.
Emotional state A yields one choice, while emotional state B yields
another. Still, which choice reflects the agent’s true preferences?83
It might be that we could ascertain the true preferences on the
basis of the absence or presence of subsequent regret. Consider that
indulging in a pleasure as a result of a hot state may lead to a feeling
of regret in the form of guilt—but guilt itself is a hot state.84 It can
and does pass. In the longer run, however, the individual may be
relieved that he has not missed out on the pleasures of life. Which is
the correct standpoint for the paternalist to adopt: the avoidance of
immediate feelings of guilty regret after the indulgence, or the later
avoidance of wistful regret over missed pleasures? The first presumes
that the initial hot state (sexual arousal or hunger) distorts true
preferences; the second presumes that the subsequent hot state (guilt)
distorts true preferences. A reasonable case could be made for
adopting either of these perspectives.85
Thus there are two preference sets to choose from, and again, no
basis by which to choose except perhaps the paternalist’s own
preferences. And if we allow the existence of an interaction between
the state of the agent during choice and his experience of the
consequences of the choice, there may be more than two preference
sets for the paternalist policymaker to choose from. As George
Loewenstein recognizes,
[I]t would clearly be suboptimal to make decisions that ignore
visceral factors. Visceral factors do affect the marginal utility of
different activities: eating is more pleasurable when one is hungry,
and sex is more pleasurable when one is aroused. . . . Clearly,
welfare maximization lies somewhere between the two extremes of

83. If we interpret this situation as a conflict of multiple selves (hot self/cold self) then
taking sides is arbitrary. See B. Douglas Bernheim & Antonio Rangel, Addiction and Cue-
Triggered Decision Processes, 94 AM. ECON. REV. 1558, 1572 (2004) (“Under that
interpretation, our use of cold preferences as a welfare standard is arbitrary.”).
84. See Kivetz & Keinan, supra note 21, at 280.
85. In the end, the application of the regret criterion is an empirical matter.
Unfortunately, there has not been very much research on the pattern of regret consequences of
actions.

930
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

making decisions that ignore visceral factors and treating visceral

influences as no different from any other influence on tastes.86
Thus even if we were to assume unambiguous post-decision regret,
this must be balanced against heightened enjoyment during or
immediately following the decision. The paternalist policymaker is
therefore faced with deciding the correct balance of the preferences
corresponding to hot and cold states.87
One possible response is to ask which preference set actually
leads to greater long-run happiness.88 Perhaps “cold” states typically
lead to choices that produce more actual happiness. Here, the
philosophical issues that we hoped to set aside in this Article become
impossible to bracket. How shall we measure actual happiness? Is it
physical pleasure, as a hedonist would suggest? Or do the agent’s
other values also come into play? And if the latter—as seems most
plausible to us—then how heavily should those values be weighed
against physical pleasure (which is surely relevant even if not
decisive)? Emotional states A and B may simply correspond to
different relative weights attached to physical pleasure and other
values—and again, theory gives us no means of determining which of
these sets of weights, or which combination of these weights,
corresponds to the agent’s “true” preferences.89
In each case, the paternalist has to decide which among equally
viable candidates to designate as the true preferences that will be
privileged by policy.

86. Loewenstein, Emotions in Economic Theory, supra note 36, at 429.

87. The complexity of this problem has been recognized by Loewenstein and
O’Donoghue, who recognize that no clear normative standard can come from this “dual-
system” analysis. See George Loewenstein & Ted O’Donoghue, Animal Spirits: Affective and
Deliberative Processes in Economic Behavior 19–21 (May 2005) (unpublished manuscript,
available at http://www.arts.cornell.edu/econ/edo1/will.pdf). The paternalist must
distinguish, on the one hand, the rational adaptation to the unconscious input of the affective
system, that is, to the tacit personal knowledge of the kinds and sources of the individual’s
well-being, and, on the other hand, the “excessive” yielding to affective demands because of
limited willpower. Id. at 38.
88. Is the total undiscounted amount of lifetime happiness the relevant standard? On
the other hand, if we must discount, then the issues of the previous section on discount rates
reassert themselves.
89. One of the factors that enables Bernheim and Rangel to rationalize their use of cold
preferences as the welfare standard is the assumption that the individual simply seeks to
maximize discounted hedonic utility. See Bernheim & Rangel, supra note 83, at 1572 (“Since
the individual has only one set of [true] preferences, discounted experiential utility . . .
accurately measures his well-being, and is unambiguously the appropriate welfare standard.”).

931
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

VII. DISCOVERING THE EXTENT OF DECISION-MAKING BIAS

It is not enough to know that a bias exists. Nor is it enough to
have identified a baseline of “true” preferences. The paternalist
policymaker also needs to know the extent of the bias in order to
design the appropriate solution to counteract it. This is not an easy
task. Numerous problems arise in this process. For the remainder of
this section, we will assume for the sake of argument that the
problem (discussed in Part I) of identifying “true” preferences has
somehow been resolved.

A. Lack of Precision in Measuring Extent of Bias

Precision in measuring the extent of any given bias has
substantial policy relevance. A large bias will justify some policies that
a small bias will not. The size of the bias will also affect the optimal
degree of intervention. Excessive intervention can reduce welfare
below pre-intervention levels. We will focus on four illustrative biases
whose extent matters for policy: hyperbolic discounting, status quo
bias, hot and cold states, and optimism bias.

1. Hyperbolic discounting
In order to craft wise policies to correct problems created by
present-bias, it is necessary to determine the extent of present-bias. If
people need encouragement to save more, the extent of their
present-bias will affect how much encouragement they require. The
optimal size of a fat tax depends on the extent of present-bias in
eating choices. Only after determining the extent of present-bias in
the areas they wish to regulate could paternalists suggest a possible
solution to counter present-bias.
Unfortunately, “[t]here is extraordinary variation across studies,
and sometimes even within studies” in estimates of intertemporal
discount factors.90 Even when the same data set is analyzed using
different, but standard, econometric techniques, there is often large
variation in discount estimates.91 Given the current technology of
estimation, the “spectacular disagreement among dozens of

90. Frederick, Loewenstein & O’Donoghue, supra note 22, at 393.

91. See, e.g., id. at 385 (using an example with a range of discount rates between 1% and
14%).

932
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

studies”92 implies an even greater variation in the predicted welfare

effects of different policies based on “correcting” rates of excessive
impatience. As we see below, even rather small differences in
attributed discount factors can be associated with significant
differences in welfare. We discuss two examples.
In an important Brookings study on savings for retirement,
David Laibson and coauthors seek to evaluate the welfare impact of
tax-deferred defined contribution retirement saving (DC) plans to
consumers with self-control problems.93 Such consumers tend to
under-save because, although they recognize the “true” value of
savings as between periods 2 and 3 and all other delayed pairs, they
are excessively impatient as between the present and the next period
(that is, they engage in quasi-hyperbolic discounting). As a
consequence, individuals continually plan to save in the future,
according to their true preferences, but always fall short of their
goals when the time arrives. Therefore, these individuals, if
sophisticated enough to correctly forecast their lapses from optimal
discounting, would value a commitment technology that would bind
them to their plans, especially if it were costless and perfect. DC
pension plans approximate such a technology because there are
generally tax penalties for early withdrawal and because, if individuals
change (lower) their contributions, the effect in increased
consumption is somewhat delayed.94 Laibson and coauthors provide
several simulations that suggest significant differences in the welfare-
enhancing effects of making DC plans available to individuals with
different present-bias factors.95 The present bias factor is typically
represented by β, which is the additional discount applied to future
periods when they are compared to the present. According to
Laibson’s calculations, the gross value of a DC plan to a twenty year-
old high school graduate varies from 28% of his current annual
consumption if β = 1 (that is, no present-bias), to 71% if β = 0.85, to
99% if β = 0.8.96 And if β = 0.60, the impatience factor derived from

92. Id. at 389. In addition, “there is no evidence of methodological progress in that the
range of estimates does not seem to be shrinking with time.” Dilip Soman et al., The Psychology
of Intertemporal Discounting: Why are Distant Events Valued Differently from Proximal Ones?,
16 MARKETING LETTERS 347, 354 (2005).
93. David I. Laibson, Andrea Repetto & Jeremy Tobacman, Self-Control and Saving for
Retirement, 1998 BROOKINGS PAPERS ON ECON. ACTIVITY 91.
94. Id. at 144–45.
95. Id. at 145–67.
96. Id. at 165.

933
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

much of the experimental evidence, the linear extrapolations of these

simulations, suggest that “the true hyperbolic [excessive impatience]
effect is two to three times as large as the effects reported
above . . . .”97 These results indicate that the desirability of policies
to encourage more savings depends crucially on the extent of bias.
Moreover, the appropriate amount of encouragement (i.e., how
large the tax penalty for early withdrawal should be) will also depend
on the extent of bias. Too much encouragement can cause a
departure from the standard of true preferences that is as great as or
greater than too little encouragement.98
Now we turn our attention to optimal sin taxes. O’Donoghue
and Rabin develop a model in which consumers choose between a
composite good and a “sin good” defined as an immediately
enjoyable good with longer-run bad health consequences.99
O’Donoghue and Rabin simulate optimal taxes for plausible values
of the other relevant parameters, as both the proportion of the
population with self-control problems and the extent of their self-
control problems vary. For example,
[i]f half the population is fully self-controlled while the other half
the population has a very small present bias of β=0.99, then the
optimal tax is 5.15%. If instead the half the population with self-

97. Id.
98. Accurately ascertaining the extent of the impatience bias is also important in
assessing the welfare impact of Social Security on naïve agents—that is, agents who have self-
control problems of which they are unaware or which they forecast incorrectly. One argument
for compulsory Social Security is that such individuals will, if left to themselves, save less than
they “really” want and thus have a lower than optimal retirement income. Ayse İmrohoroğlu
and coauthors conclude, based on their simulations, that Social Security does not raise welfare
from the perspective of almost any age for individuals with impatience factors in the
neighborhood of 0.85 to 0.90. Ayse İmrohoroğlu, Selahattin İmrohoroğlu & Douglas H.
Joines, Time-Inconsistent Preferences and Social Security, 118 Q.J. ECON. 745, 781 (2003).
This is because an unfunded retirement scheme, such as Social Security, lowers the aggregate
capital stock and thus income at all ages. Id. at 770. While Social Security redistributes existing
income to those in retirement, “the utility gains from increased old-age consumption are too
small to offset the losses from reduced consumption earlier in life.” Id. at 776. However, all
this changes, as may be expected, when the degree of impatience increases. Under those
circumstances, the amount of under-saving may be so great that the increase in income during
retirement brought about by Social Security payments will swamp the effects of a lower
aggregate capital stock. In fact, the simulations reveal that “[s]ocial security does significantly
raise welfare with β = 0.60 . . . .” Id. at 781. Obviously, government policies regarding savings
for retirement will be affected substantially by the extent of the impatience bias. A relatively
small bias may suggest the substitution, in whole or part, of fully-funded or private retirement
plans for the current Social Security scheme.
99. O’Donoghue & Rabin, Optimal Sin Taxes, supra note 3.

934
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

control problems has a somewhat larger present bias of β =0.90—

which is still a smaller present bias (larger β) than often discussed in
the literature—the optimal tax is 63.71%.100
Therefore, if the government were to estimate the present bias as the
latter (lower β) when in fact it was the former, it would reduce
consumer welfare by imposing a tax about twelve times too large.
Notice also that O’Donoghue and Rabin’s approach includes
assumptions about what percentage of the population is afflicted by
present-bias101—an issue we will address more fully later.
The difficulty for policy prescriptions is that no one is very
confident about the true impatience or present-bias factor (β), nor
about the proportion of the population subject to the bias. This is
one reason that, in the various studies discussed above, the authors
show the effects on welfare for various calibrations of the relevant
parameters. In sum, the extent of the impatience bias is very
significant in determining whether a specific paternalist policy
increases or decreases welfare relative to the status-quo. Current
estimates are unable to provide a basis for policy prescriptions that
reliably increase welfare. At best, policies derived from the current
state of knowledge can only produce certain objective results, like
more saving or lower junk-food consumption, that may or may not
increase welfare. Therefore, the new paternalism, supposedly based
on the underlying normative preferences of individuals, shades into
the old paternalism, based on what is “objectively best” in the
opinion of an outside observer.

2. Status quo bias

If default savings plans are justified on grounds of status quo
bias, then we need a measure of the extent of that bias. The greater
the status quo bias, the more the selected default savings plan
matters because more people will stay with it longer.102 If that plan is
not optimal, then individuals will be stuck in a relatively low-welfare
savings outcome. How bad this situation is and how long people will

100. Id. at 1838.

101. Id.
102. Status-quo bias is usually estimated by the (disproportionate) frequency with which
the status-quo option is accepted by decision-makers. The duration of the bias—how long
people stay with the option—has not been systematically measured.

935
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

be in it depend on the welfare losses from suboptimal savings and

the extent of the bias.103
In their seminal study, Samuelson and Zeckhauser found that the
status-quo bias, even where statistically significant, differs in size
across tasks and alternatives—from substantial effects to small
effects.104 Whether this is due to systematic contextual factors, the
inherent variability of the phenomenon, or difficulties in
measurement technique, is impossible to say at this time.
Furthermore, whether these effects, regardless of their magnitude,
are caused by rational transaction cost factors or behavioral biases is
difficult to determine.105
Moreover, Samuelson and Zeckhauser found that the bias is
larger when the individual’s preference for a neutrally-presented
alternative to the status-quo is weaker.106 Thus, the size of the bias is
likely to depend on the default; if people know they are already
saving something by default, they may be less likely to take the time
to change to a better plan. Because of the generally low returns to
the default allocations, Choi and coauthors found that automatic
enrollment produced offsetting effects: “While higher participation
rates promote wealth accumulation, the low default savings rate and
the conservative default investment fund undercut accumulation,”
and, in their sample, the two effects were approximately equal in
magnitude.107 So, in the aggregate, these individuals were in no
better position than before. On the other hand, the farther the

103. The issue here is somewhat more complex because what we must know is optimal
401(k) savings, since people save in other ways. This is only indirectly related to the general
rate of excessive impatience.
104. Samuelson & Zeckhauser, supra note 30, at 15–17 tbls.1a, 1b & 1c. The absolute
size of the status-quo bias is SQ-NEUT and the relative size is (SQ-NEUT)/NEUT where SQ
is the choice frequency for a given alternative when it is in the status-quo position and NEUT
is the frequency when the alternative is presented neutrally. Id. at 15–17.
105. In their analysis of the impact of status-quo bias on decisions regarding enrollment
in 401(k) programs, Madrian and Shea observe, “Unfortunately, there is no way to disentangle
the magnitude of rational, transaction costs motivated procrastination from behavioral, self-
control motivated procrastination in the data.” Madrian & Shea, supra note 80, at 1180. They
do note, however, that there is a “possibility of the latter.” Id.
106. They state their equivalent conclusion in terms of the converse proposition: “The
stronger was an individual’s preference for a selected alternative, the weaker was the bias.”
Samuelson & Zeckhauser, supra note 30, at 8. See also James J. Choi, David Laibson, Brigitte
Madrian & Andrew Metrick, Optimal Defaults, AM. ECON. REV., May 2003, at 180, 183–84.
107. James J. Choi, David Laibson, Brigitte Madrian & Andrew Metrick, For Better or
For Worse: Default Effects and 401(k) Savings Behavior 2 (Pension Research Council, Working
Paper No. 2002-02, 2002).

936
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

default is from the individual’s optimum savings rate, the greater the
probability that he will opt out of the default and begin saving
optimally. A sufficiently inappropriate default will weaken the status-
quo bias and motivate change.108 In this case, the paternalist should
not be searching for a welfare-enhancing default, but for one that is
far enough from it to encourage active decision-making. Therefore,
the nature of the paternalist’s task will depend not only on knowing
the relevant size, persistence, and cause of the status-quo bias, but
also its responsiveness to the alternatives considered. Neither the
economist nor the paternalist has adequate measures of any of these
factors.109

3. Endowment effects
The size of the endowment effect clearly determines whether any
paternalistic change in the assignment of default contractual rights
can increase welfare. If endowment effects are weak or even
nonexistent, then even if the paternalist selects the optimum rights
package, no purpose is served by presuming vacation time, dismissal
only for cause, etc., in employee contracts beyond saving on
transaction costs. Of course, if the paternalist does not select the
optimum rights package, transaction costs will be increased.
Until recently, the behavioral literature accepted the existence of
endowment effects without much controversy. Surprisingly, the
existence of these effects has never been adequately tested. Kathryn
Zeiler and Charles Plott have undertaken and reported experiments

108. See Choi, Laibson, Madrian & Metrick, Optimal Defaults, supra note 106, at 183–
84; see also Samuelson & Zeckhauser, supra note 30, at 8 (“The stronger was an individual’s
preference for a selected alternative, the weaker was the bias.”).
109. Another superficially attractive possibility is to choose a default that minimizes the
total realized costs of opting out. However, this is not a welfare-maximizing or enhancing
standard in the presence of status-quo bias. As we have seen above, a default that motivates
people to abandon more rapidly their suboptimal savings rate may be a good thing. This
implies that the correct standard is the minimization of the sum of the realized costs of opting
out and the flow losses due to too little or too much savings. In other words, higher realized
costs of opting out would in fact be welfare enhancing if they were accompanied by a larger
reduction in the costs of nonoptimal savings. But see RICHARD H. THALER & CASS R.
SUNSTEIN, NUDGE: IMPROVING DECISIONS ABOUT HEALTH, WEALTH, AND HAPPINESS 109
(2008) (claiming that a low rate of opting out under an automatic enrollment default is
welfare-enhancing because it reveals that people are in a better position). Individuals could stay
in that position simply because they have not been sufficiently motivated to choose a more
nearly optimal savings plan, that is, because they are experiencing the very status-quo bias that
Thaler and Sunstein view as an important cause of suboptimal savings.

937
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

that control for the most important factors that may be responsible
for the appearance of endowment effects.110 When factors that could
plausibly affect the nature of the good in question (such as whether
it is perceived as a gift from the experimenter which could be
impolite to exchange, or whether the endowment is really a signal of
private information from the experimenter or perhaps other subjects)
are eliminated, the results suggest “[e]ither no ‘endowment effect’ of
the sort predicted by prospect theory exists [in these experiments] or
the effect is sufficiently weak that other phenomena easily swamp
it.”111 But notwithstanding these results, let us suppose that true
endowment effects exist. Accurate measurement of their magnitude
will determine the efficacy of new default rules in improving welfare.
If the effects are small, as Zeiler and Plott suggest, then default rules
will simply increase transaction costs for people to return to their
favored packages. The more difficult the default is to escape, the
greater will be the resulting loss.

4. Hot and cold states

To create an optimal policy justified on the basis of hot-state
bias, such as a cooling-off period, policymakers need to know how
much people are affected by their hot states; that is, to what extent
their decisions are distorted. It is possible they may not be distorted
at all. The process by which they are supposedly distorted is through
the “empathy gap”—the tendency of individuals in a hot or excited
emotional state to overestimate the intensity of the hedonic
consequences of an event or good later on when the hot state has
dissipated.112 However, hedonic consequences are not the sole, and,
in many cases, not even the primary determinants of choice, as when
people sacrifice personal pleasure to send their children to college or
to pursue some form of excellence. Thus, overestimation of hedonic
consequences may not have a significant impact on many decisions.

110. Charles R. Plott & Kathryn Zeiler, Exchange Asymmetries Incorrectly Interpreted as
Evidence of Endowment Effect Theory and Prospect Theory?, 97 AM. ECON. REV. 1449 (2007).
This study stands out as the most rigorous attempt to date to control for confounding factors.
See id. at 1454–56.
111. Id. at 1463.
112. See Timothy D. Wilson & Daniel T. Gilbert, Affective Forecasting, 35 ADVANCES
EXPERIMENTAL SOC. PSYCHOL. 345, 365–66 (2003). It is also the case that when in a cold
state people underestimate the intensity of feelings in a hot state. Id.

938
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

Furthermore, hotness and coldness refer to the state of the

affective system. But even this is a simplification of the problem,
because affect is not just hot or cold. Affect is an important element
in the decision-making process; it comprises the individual’s
motivational system—hunger, thirst, and the desire to have sex are
just a few examples. The affective system suggests the options an
individual must consider. Then it reduces or flags these options in
accordance with the individual’s specific goals, time frames, and
perhaps most importantly, his acquired knowledge of local
circumstances.113 For example, a pleasant or unpleasant feeling may
follow from thinking about meeting an old acquaintance or thinking
about consuming a product with particular mental associations.
These images are marked by a somatic state.114 So an option that is
likely to result in such a meeting or consuming such a product may
immediately, that is, without much deliberation, be chosen or
discarded by the individual as a result of the somatic state.
Consequently, a rational choice can seem, at times, as if it were
choice without reasoning. We might see an individual jumping to
conclusions, hastily eliminating alternatives and making decisions,
and “wanting” things without a “sufficient” reason.115 None of this
suggests that choices are distorted; this is the way human beings
choose—a combination of explicit deliberation and affect. Affect is
part of rationality.
Of course, it is possible for the affective system to break down
and to produce distorted choices.116 If we allow that this will happen
in some cases, what must the policymaker know? First, he must be
able to distinguish distortive affect from normally-functioning affect.
As we have seen, appearances do not suffice. Second, even if the

113. ANTONIO R. DAMASIO, DESCARTES’ ERROR: EMOTION, REASON AND THE HUMAN
BRAIN 173, 181–83 (1994).
114. The body states produced by processes of the affective system necessary for rational
decision-making may be conscious or unconscious, that is, they may or may not constitute
feelings. Body states may be activated by stimuli but not be the focus of awareness or attention.
Nevertheless, they can affect “cognitive processes in a covert manner and thus influence the
reasoning and decision-making mode.” Id. at 185.
115. This is particularly likely in the case in which the individual himself will not
experience a feeling of liking, hating, fearing, and so forth. He will simply approach or avoid,
want or not want, without explicit liking or disliking.
116. See, e.g., Kent C. Berridge, Pleasure, Unfelt Affect, and Irrational Desire, in
FEELINGS AND EMOTIONS: THE AMSTERDAM SYMPOSIUM 243, 254–59 (A.S.R. Manstead,
N.H. Frijda & A.H. Fischer eds., 2004) (using the example of irrational choice arising from
addictions).

939
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

policymaker knows that a certain type of affective state can be

distortive, he must know to what extent it distorts in a particular
context. Small distortions are hardly worth policy attention. Third,
he must know something about the rate at which this particular
distortive hot state dissipates. Cooling-off periods should be
calibrated to this rate. A longer cooling-off period would impose
excess costs on the sellers and then to the consumers. And, finally, he
must know the degree to which ex post rationalization of a decision
that cannot be changed will obviate the need for cooling-off
periods.117

5. Optimism and availability bias

To create an efficient policy designed to counteract optimism
bias, such as requiring risk narratives for risky products or
investments, policymakers need to know both the extent of people’s
excessive optimism (how much they underestimate the risk) and the
extent of their availability bias (how much they respond to a
narrative with varying levels of scariness). More excessive optimism
points toward more and scarier narratives, greater availability bias
toward fewer and less scary narratives. But what is the standard by
which the paternalist’s policy will be judged? Presumably, he wants
the two biases to balance at the point where people make the rational
decision. It would not be sufficient to know the “correct” estimate
of risk (and then to try to induce this perception by the appropriate
narrative), because how much risk a person ought to bear is
independent of neither his subjective attitude toward risk nor his
subjective assessment of the cost of the bad outcome associated with
it. The paternalist needs some indication of, say, the right product to
buy or the right mutual fund in which to invest. However, if he
knew this, then he would not need to worry about offsetting

117. Wilson and Gilbert argue that cooling-off periods might actually make people less
satisfied with their decisions because they inhibit the process of rationalization. Timothy D.
Wilson & Daniel T. Gilbert, Affective Forecasting: Knowing What to Want, 14 CURRENT
DIRECTIONS PSYCHOL. SCI. 131, 133 (2005) (“When people make a decision that is difficult
to reverse, such as buying a sweater from a store with a ‘no returns’ policy, they are strongly
motivated to rationalize the decision and make the best of it. When people can more easily
undo a decision, such as buying a sweater they can return, they are less motivated to rationalize
their choice, because they can always change their minds. Consequently people are often happier
with irrevocable choices because they do the psychological work necessary to rationalize what
they can’t undo.” (emphasis added)).

940
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

optimism bias with appropriate risk narratives, because policy could

simply command the correct outcome.

B. Absence of a Single Measure of Bias, Even Intrapersonally

To make matters more difficult, a single measure of any given
bias generally does not exist, even for a single individual. Different
degrees of bias will exist depending on the choice situation. Here we
focus on hyperbolic discounting and hot-and-cold states.

1. Hyperbolic discounting
As discussed earlier, people’s actual behavior in situations of
intertemporal choice appears to approximate hyperbolic (not quasi-
hyperbolic) discounting. This means that there is no single factor β
that represents the agent’s degree of present-bias. Instead, the extent
of bias depends on the distance between the two future periods
compared and the present.
Suppose the paternalist policymaker has (somehow) determined
that the agent’s true preferences are best represented by some fixed
discount factor δ. If the agent’s actual behavior approximates
hyperbolic discounting, then the agent will discount the future too
much when comparing periods relatively close to the present. But
what is the extent of that bias? The answer will depend on how close
the two periods compared are to the present. The closer they are to
the present, the greater will be the present-bias. On the other hand,
the agent will also discount the future too little when comparing
periods sufficiently far from the present. This is a necessary result of
the paternalist’s having designated a fixed discount factor δ as
correct, when actual behavior reflects discount factors both higher
and lower than δ. This conclusion could only be avoided by the
paternalist having assumed the correct discount factor is the highest
(most patient) one the agent ever exhibits.
The implication for policy is that bias-correcting policies should
be calibrated to the distance from the present of the intertemporal
decisions being made. Take, for instance, a fat tax designed to curb
junk-food consumption. The tax would need to be higher when a
person is buying food for immediate consumption—say, at a
restaurant or convenience store. The tax would need to be lower
when a person is buying food for more distant consumption—say, at
a grocery store. And depending on the policymaker’s judgment
about the correct amount of discounting, it might even be necessary

941
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

to subsidize fat, rather than taxing it, for very distant consumption—
say, when planning for a celebration a year from now (like a
wedding). Again, this follows directly from the selection of a single
correct discount factor in the context of hyperbolic discounting.
Even if the policymaker has selected an extremely high discount
factor so that no subsidies are required, he still needs to make the tax
a function of the degree of present-bias that applies to any given time
frame, which requires the policymaker to have knowledge of that
present-bias.
Obviously this is impractical. In reality, the policymaker would
most likely adopt a single tax that would apply regardless of context
or time frame. Such a tax would yield problems of both under-
correction (the tax would be too low for decisions close to the
present) and over-correction (the tax would be too high for
decisions far from the present). Since any change in the tax will tend
to produce more of one problem and less of the other, the
policymaker will have to weigh these effects against each other to
decide the best tax—and again, that requires having knowledge of
the actual extent of present-bias for different time frames.

2. Hot and cold states

The impact of the hot-state bias on decision-making depends on
the intensity of the relevant visceral factors.118 There are degrees of
anger, fear, hunger, or sexual desire. It stands to reason that more
intense emotions will distort decisions more than do less intense
ones.119 The degree or intensity of visceral factors depends on the
context of the decision. No decision defined in objective terms, such
as whether to marry or buy a car, is necessarily a hot decision.
Whether it is and whether it produces suboptimal choices “depends
on a wide range of influences.”120 These include “how recently a
drive was satisfied and on the presence of arousing stimuli” as well as
“the interaction of situational factors and construal processes and on
internal psychobiological factors.”121 Although some types of
decisions are no doubt more likely to be affected by visceral states,

118. George Loewenstein, Out of Control: Visceral Influences on Behavior, 65

ORGANIZATIONAL BEHAV. & HUM. DECISION PROCESSES 272, 273 (1996).
119. Id.
120. Id. at 281.
121. Id.

942
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

there will be significant intra-individual variation depending on the

particular context. As a result, waiting periods for decisions to take
effect or options to allow people to revoke prior decisions will have
different consequences as the context varies. The net consequences
may be costly in some circumstances and beneficial in others.122
Whether a general rule, even adapted to particular types of situations,
is beneficial overall requires knowledge of the relative frequency of
the relevant contextual factors—knowledge no one has.
Furthermore, the presumably different rates of hot-state dissipation
(derived from the differential presence of contextual factors) will also
determine the optimal length of the cooling-off period.123

VIII. ACCOUNTING FOR SELF-DEBIASING

People have numerous means at their disposal to mitigate the
effects of their own biases. We will refer to these methods as “self-
debiasing” or “self-regulation.” We do not claim, however, that self-
regulation effectively eliminates all or even most biases. Our
argument is rather that the existence of such methods implies that
some paternalistic policies that appear desirable at first blush are
either unnecessary or in need of softening (lower sin taxes, shorter
cooling-off periods, etc.) to account for the extent to which the
biases have already been addressed privately. Policy measures that do
not take account of self-debiasing can move the individual even
farther away from his optimal decision than he would be in the
absence of such policies.

A. The Many Varieties of Self-Debiasing and Self-Regulation

The most obvious form of self-regulation is simply the exertion
of willpower. But in an important sense, willpower comes into play
too late. When the individual is already exposed to a temptation,
direct resistance can be very costly. Individuals, however, are more
inventive about the methods they choose to achieve their long-run

122. In those cases in which the distortive aspect of hot decisions is small, the delay or
option-to-revoke costs will outweigh the benefits.
123. Sometimes the hot state is caused by contemplation of the decision itself such as
those relative to death, disease, accidents, and terrorism. Therefore, the hot state will not
dissipate so long as the decision is ultimately made. In these cases neither delay nor option-to-
revoke seems to have any paternalistic benefits. So the net result of having such cooling-off
periods is costly. See Jeffrey A. Blumenthal, Emotional Paternalism, 35 FLA. ST. U. L. REV. 1,
61–62 (2007).

943
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

objectives. Self-regulation “consists of a wide range of cognitive and

motivational operations, such as acting quickly to take opportunities,
ignoring distractions, acting flexibly in response to situations,
overcoming obstacles, and managing conflicts between goals.”124
More specifically, self-regulation functions to reduce the impact of
behavioral biases by using strategies that are cognitive,
environmental, and directly behavioral.125 Cognitive strategies
include focusing on the benefits of reaching one’s goal, distracting
oneself from undesirable behavior by using imagery of better
alternatives, and using self-praise to commend oneself for achieving
an important goal.126 Environmental strategies include avoiding
people, situations, settings, and even times of day when temptations
are strong.127 Directly-behavioral strategies include increasing social
support, utilizing cues about one’s important goals, rewarding
oneself for desirable behavior or punishing oneself for undesirable
behavior, and creating ways to make desirable behavior itself more
enjoyable.128
The following is a partial list of debiasing strategies in each
general category. The large variety of these strategies and their
connection to particular circumstances of time and place should
make obvious the scope of the difficulties paternalists face in trying
to account for them in the determination of welfare-enhancing
policies.

1. Cognitive strategies
a. Resolutions and commitments. These mental devices focus a
person’s attention on those situations and choices in which his own
biases are most likely to be manifested. A person who suffers from
weakness of will when it comes to eating might make a resolution
never to eat desserts except on special occasions. A person with a

124. Gráinne M. Fitzsimons & John A. Bargh, Automatic Self-Regulation, in

HANDBOOK OF SELF-REGULATION: RESEARCH, THEORY AND APPLICATIONS 151, 151–52
(Roy F. Baumeister & Kathleen D. Vohs eds., 2004) (citing P.M. Gollwitzer & G.B.
Moskowitz, Goal Effects on Action and Cognition, in SOCIAL PSYCHOLOGY: HANDBOOK OF
BASIC PRINCIPLES 361, 368 (E.T. Higgins & A.W. Kruglanski eds., 1996)).
125. For a list of self-regulatory strategies, see ENCYCLOPEDIA OF MENTAL DISORDERS,
Self-Control Strategies, http://www.minddisorders.com/Py-Z/Self-control-strategies.html.
126. Id.
127. Id.
128. Id.

944
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

marked tendency to make rash decisions in hot emotional states—

say, when confronted with the opportunity to commit adultery—
might resolve to physically remove himself from the situation before
making any decision or to count slowly to ten before taking any
action.
b. Mental accounts and budgets. As a means of setting boundaries
and reminding themselves of their resolutions, people sometimes
adopt mental accounting devices to keep certain behaviors within
limits.129 For instance, a person might establish an entertainment
budget; he allows himself to spend as much as he likes on
entertainment up to a chosen limit, but will not let himself exceed it.
Or he might establish a fund for household expenses that cannot be
tapped for other purposes. Mental budgets can enable indulgence as
well as limit it, such as when someone commits himself to take a
vacation (perhaps to overcome a tendency toward overworking).

2. Environmental strategies
a. Submission to social controls. These are efforts to enlist
outsiders to assist in the keeping of one’s commitments. Someone
trying to quit smoking may advertise that intention to friends and
family, so they will remind him of his commitment and frown on
deviations from it. Formal organizations like Alcoholics Anonymous
and Weight Watchers play the same role, providing a support
network that lowers the cost of following commitments and raises
the cost of breaking them. Strotz provides the more extreme
examples of getting married “for the sake of ‘settling down’” or
joining the army as methods of precommitting financial or economic
actions.130
b. Self-constraining devices. These devices structure the external
environment to raise the cost of some activities and lower the cost of
others. People trying to quit smoking sometimes throw away their
cigarettes to remove the temptation. People with eating problems
may refuse to allow especially tempting foods in their home. People
who have difficulty saving can opt to have automatic monthly

129. That people use mental budgeting to control their behavior is well established. See,
e.g., Chip Heath & Jack B. Soll, Mental Budgeting and Consumer Decisions, 23 J. CONSUMER
RES. 40 (1996); Richard Thaler, Mental Accounting and Consumer Choice, 4 MARKETING SCI.
199 (1985); Klaus Wertenbroch, Consumption Self-Control by Rationing Purchase Quantities of
Virtues and Vice, 17 MARKETING SCI. 317 (1998).
130. Strotz, supra note 25, at 173.

945
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

transfers from their checking accounts to their savings accounts.

Gamblers can limit their opportunities to exceed self-imposed limits
by leaving their credit and ATM cards at home.

3. Directly-behavioral strategies
a. Internal rewards and punishments. It is not uncommon for
people to affect their choices by means of internally imposed
incentive schemes, by which they give themselves rewards for more
favored behavior and punishments for less favored behavior. For
example, someone trying to lose weight might reward herself for
meeting weight-loss goals with permission to go to a movie or buy
another music CD. The phenomenon of self-gifting has been
documented in a series of papers,131 and the efficacy of self-reward
schemes in motivating greater effort and performance has also been
shown.132

B. The Significance of Context for Self-Regulation

Consideration of the various methods used by real people to
regulate their own behavior reveals the overriding importance of
context.133 Resolutions, commitments, mental budgets, and internal
rewards and punishments typically depend for their application on
specific features of time and place: what time of day it is; whether
one is at work, at home, or on vacation; whether the present
situation is a special occasion like a birthday or wedding; and so on.

131. See David Glen Mick, Self-Gifts, in GIFT-GIVING: A RESEARCH ANTHOLOGY 99

(Cele Otnes & Richard F. Beltramini eds., 1996); David Glen Mick & Michelle DeMoss, Self-
Gifts: Phenomenological Insights from Four Contexts, 17 J. CONSUMER RES. 322 (1990).
132. See Albert Bandura & Dale H. Schunk, Cultivating Competence, Self-Efficacy and
Intrinsic Interest through Proximal Self-Motivation, 41 J. PERSONALITY & SOC. PSYCHOL. 586,
586–87, 595–97 (1981); Albert Bandura & Bernard Perloff, Relative Efficacy of Self-Monitored
and Externally Imposed Reinforcement Systems, 7 J. PERSONALITY & SOC. PSYCHOL. 111, 111,
114–116 (1967).
133. All actions derive their meaning from context. Consider the simple act of an
individual touching his nose with a finger. The meaning changes with the context in which it
takes place. If the individual is being asked to touch his nose as part of an experiment, the
purpose of which is unknown to him, then he may see the act as simply obeying the
instructions for the sake of science or for some payment. If it is a part of a neurological exam,
then the context is the health of the individual or the diagnosis of a possible disease.
Alternatively, it may simply be scratching an itch or swatting a fly. Or it may be a socially-
recognized gesture of disapproval. Context determines meaning. For a detailed analysis, see
Shaun Gallagher & Anthony J. Marcel, The Self in Contextualized Action, 6 J.
CONSCIOUSNESS STUD. 4 (1999).

946
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

Self-constraining devices and submission to social controls are

devices designed to affect one’s context by inserting costs, benefits,
barriers, and reminders that would not otherwise be present.

1. Self-regulation in the laboratory versus in the wild

As has been said, the existence of self-debiasing measures does
not mean that all biases are perfectly corrected. Individuals may not
know all of their biases, and the methods they adopt may not
succeed—or may succeed too well, as in the case of anorexics or
tightwads. The point is that any debiasing policy will only be
successful to the extent that it takes self-debiasing efforts into
account. Measuring the extent of bias absent self-debiasing efforts
will overstate the degree of bias realized in behavior, and hence the
amount of correction required. Without a clear and realististic
context, laboratory measurement of self-regulation or control will
not capture this.
To see why this should matter to the paternalist policymaker, let
us take the case of an individual who faces future consequences to be
balanced against current costs or benefits. His actual rate of
impatience is greater, we assume, than what would be dictated by his
“true” preferences.134 Let us suppose, charitably, that the
policymaker already knows the individual’s true time preference, as
well as the unmodified extent of his present-bias (that is, excessive
impatience). Nevertheless, he still needs to know the degree to
which the individual’s self-regulatory mechanisms counteract his own
impatience. This would give the policymaker an effective or
operational level of present-bias, which will—if the person’s self-
regulation works at all—differ from his unmodified present-bias. This

134. Not all self-regulation or “self-control” problems are impatience problems. There
can be a negative difference between the actual rate and the normative rate. See, e.g., John
Ameriks, Andrew Caplin, John Leahy & Tom Tyler, Measuring Self-Control 1 (Nat’l Bureau of
Econ. Research, Working Paper No. 10514, 2004) (“The current view of self-control
problems as involving the need to suppress the immediate urge to consume is inadequate. In
our sample, ‘present-bias’ (the urge to consume today more than would be ideal) is no more
prevalent than is ‘future-bias’ (a tendency to consume less today than would be ideal) . . . .”).
The individual can have “excessive patience,” as when he operates under the comforting
illusion that he will not die or grow infirm or be less capable over time of enjoying physical
activity. See Wojcieh Kopczuk & Joel Slemrod, Denial of Death and Economic Behavior, 5
ADVANCES THEORETICAL ECON., Jan. 2005, art. 5, at 2–4, available at http://www.
bepress.com/bejte/advances/vol5/iss1/art5. We ignore this here for purely heuristic reasons,
but caution that this hyperopia can complicate policymakers’ decisions.

947
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

knowledge is necessary to determine, for example, the rate of

taxation on present benefits that will lower the effective rate of
impatience to the correct level.
To solve the problem, the paternalist needs to measure the
amount of self-debiasing that occurs, but doing so is inherently
problematic. Much of our evidence of decision-making biases derives
from laboratory experiments. But laboratory experiments cannot
capture all debiasing efforts, because self-debiasing efforts are
context-dependent: the person with a weight problem resists desserts
except on special occasions, the person trying to save more money
signs up for automatic deductions from paychecks (but not
unexpected windfalls), etc. Lab environments, on the other hand, are
typically devoid of context. Even if the experiment designer
deliberately structures the experiment to create the illusion of
context, this effort cannot capture self-debiasing efforts that seek to
achieve overall outcomes by differing across contexts. What may
appear to be a bias in a particular context could be part of an overall
plan that creates a deliberate exception in that area. The strategy of
eating dessert only on special occasions, for example, rations fat
consumption by defining narrow contexts in which it is allowed. A
lab environment can duplicate one context, but not all the contexts
relevant to the individual’s overall strategy.
The context may be inferred by external observers of situations
when observing human behavior “in the wild,” or it may be supplied
by the observers in experimental situations. The main question for us
is whether the context supplied by observers or inferred by subjects
in an experiment is equivalent to the context of the real-world
situations to which the results of these experiments are
generalized.135 Thus, if an individual is asked as part of an experiment
whether he prefers a larger, later reward as opposed to an earlier,
smaller one over various intervals of time, he may or may not display
a range of time-discounting propensities that reflect his real-world

135. The equivalence of context has implications for both problem construal (“What am
I being asked to do?”) and for the nature of the solution (“How should I behave?”). See, e.g.,
Glenn W. Harrison & E. Elisabet Rutström, Doing It Both Ways—Experimental Practice and
Heuristic Context, 24 BEHAV. & BRAIN SCI. 413, 413–14 (2001) (“Field referents can often
help subjects overcome confusion about the task. . . . [Even] [i]n cases where the subject
understands all the relevant aspects of the abstract game, problems may arise due to the
triggering of different methods for solving the decision problem. The use of field referents
could trigger the use of specific heuristics from the field to solve the specific problem in the
lab, which otherwise may have been solved less efficiently . . . .” (citations omitted)).

948
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

behavior. This will depend on the degree of similarity in context

between the experiment and the wild.
Thus, self-regulation is context-dependent. The drive for
generality in experiments, on the other hand, usually produces
minimal or antiseptic context such as designating some actors as
buyers or sellers, determining the set of alternatives, and ordering
their presentation.136 The experimenters may worry that providing
“too much” context limits the applicability of results to the real
world. But precisely the opposite is the case if the purpose is to
inform paternalist policy. To capture the real world of choice, we
must see choices in their self-regulatory context. There is no such
thing as general or abstract self-regulation137—although the
contextual nature of self-regulation is obscured by the popular idea
that it is derived from some homogeneous source such as inner
strength or willpower.138 Thinking in this way is apt simply to result
in measuring choice propensities in under-defined contexts.

2. The automaticity of unconscious self-regulation

There is important and growing evidence suggesting that
“conscious processes are neither necessary or even typical for
effective self-regulation . . . .”139 Much self-regulation must be non-
conscious to be effective in view of the limited capacity of individuals
to deal with a complex and rapidly-changing environment in a fully
deliberative manner.140 There are, for example, unconscious
processes associated with selective attention, that is, the focusing on

136. See George Loewenstein, Experimental Economics from the Vantage-Point of

Behavioural Economics, ECON. J., Feb. 1999, at F25, F29 (1999) (“Many experimental
economists seem to view their enterprise as akin to silicon chip production. Subjects are
removed from all familiar contextual cues. . . . [B]uyers and sellers become ‘persons A and B,’
and all other information that might make the situation familiar and provide a clue about how
to behave is removed.”).
137. It is possible to make an even broader claim. See id. at F30 (“A major discovery of
cognitive psychology is the degree to which all forms of thinking and problem solving are
context-dependent . . . .”).
138. This does not seem to be the case (or, at least, the metaphor does not seem
appropriate) since there are intrapersonal differences across self-regulatory tasks and various
situations. See Daniel Cervone, People Who Fail at Self-Regulation: What Should We Think of
Them—and How?, 7 PSYCHOL. INQUIRY 40, 41 (1996).
139. Fitzsimons & Bargh, supra note 124, at 151 (providing a partial survey of the
relevant literature).
140. Id. at 152.

949
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

important or superordinate goals.141 There is also unconscious

modulation of emotional states that might threaten the attainment of
these goals.142 And subliminally activated goals have been shown to
“guide behavior in a purposive, though nonconscious,
manner . . . .”143 These unconscious processes are triggered by the
local and personal circumstances of the individual, that is, by his self-
regulatory context, to an even greater extent than the conscious
processes discussed above.144 For example, simply thinking about
people with whom one has a relationship, such as family, friends, and
colleagues, can automatically “activate goals that guide and regulate
the self’s actions in a given situation . . . .”145 These goals are
generally those congruent with the attitudes of the others.146
Social norms relevant to the particular environment in which the
individual acts are also sources of automatic processes.147 Most
interestingly, an automatic form of self-control known as
“counteractive self-control” can be triggered by imagining the
temptation. This is a proactive or ex ante adjustment of the relevant
choice variables.148 Counteractive self-control may involve changing
the “objective” choice situation by self-imposing a penalty for the
failure to achieve one’s long-term goal.149 Additionally, it may
change the psychological meaning of the choice situation by raising

141. Id.
142. The existence of mood regulation tends to counteract “irrational” pressures on
decision-making when the stakes are high. See Ralph Erber, Maureen Wang Erber & Jennifer
Poe, Mood Regulation and Decision-Making: Is Irrational Exuberance Really a Problem?, in 2
PSYCHOLOGY OF ECONOMIC DECISIONS: REASONS & CHOICES 197, 204–05 (Isabelle Brocas
& Juan D. Carrillo eds., 2003).
143. Fitzsimons & Bargh, supra note 124, at 153–55 (citation omitted).
144. See id. at 156–57 (discussing how social environment and personal relationships can
affect unconscious self regulation).
145. Id. at 157 (citations omitted).
146. Id.
147. Id. at 156.
148. Thus, counteractive self-control is not dissonance reduction. See Ayelet Fishbach &
Yaacov Trope, The Substitutability of External Control and Self-Control, 41 J. EXPERIMENTAL
SOC. PSYCHOL. 256, 259 (2005) (“CCT [Counteractive Control Theory] concerns proactive
attempts to enact what one ideally prefers, whereas dissonance concerns attempts to reduce the
discomfort produced by having failed to enact what one prefers.”).
149. See id.

950
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

the subjective value of the long-term goals and decreasing the

subjective aversion of the short-term costs.150
Unconscious self-regulation is not easily observable. The target
agent himself is unaware of its operation. It may take quite subtle
forms, as we have seen above. For the new paternalist, accounting for
this type of self-regulation is thus particularly difficult.

IX. ACCOUNTING FOR INTERDEPENDENT BIASES

The simultaneous existence of more than one bias affecting the
individual’s cognition or behavior poses a difficult problem for policy
choices grounded in the new paternalism.151 Almost universally, in
the current state of research, only one bias at a time is studied.152 But
since we have good reason to believe that simultaneous biases are
likely, merely finding a bias that is significant both statistically and in
size is not sufficient to conclude that the associated behavior is
suboptimal.153
The identification of myriad cognitive and behavioral biases
across hundreds of studies, as well as sometimes the identification of
more than one bias within a single study, is good prima facie
evidence that individuals are subject to multiple biases. Joachim
Krueger and David Funder present a “partial list” of forty-two
cognitive biases, including numerous opposite or contradictory
biases, discovered in the social psychology literature since 1985.154
The likelihood of multiple biases in individual behavior and
cognition has both a qualitative and a quantitative impact on optimal
policy.

150. For a survey of results, see Yaacov Trope & Ayelet Fishbach, Going Beyond the
Motivation Given: Self-Control and Situation Control over Behavior, in THE NEW
UNCONSCIOUS 537, 537–51 (Ran R. Hassin, James S. Uleman & John A. Bargh eds., 2005).
151. This problem should be distinguished from those arising from the existence of
multiple biases within a population. We do not deal with this here.
152. See Hanming Fang & Dan Silverman, Distinguishing Between Cognitive Biases, in
BEHAVIORAL PUBLIC FINANCE 47, 48 (Edward J. McCaffery & Joel Slemrod eds., 2006) (“So
far, both the theoretical and the empirical studies in economics have tended to investigate the
implications of cognitive biases and heuristics one bias at a time . . . .”).
153. See generally Gregory Besharov, Second-Best Considerations in Correcting Cognitive
Biases, 71 S. ECON. J. 12 (2004).
154. See Joachim I. Krueger & David C. Funder, Towards a Balanced Social Psychology:
Causes, Consequences, and Cures for the Problem-Seeking Approach to Social Behavior and
Cognition, 27 BEHAV. & BRAIN SCI. 313, 317 tbl.1 (2004).

951
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

A. Qualitative Effects
In this section we follow the analysis of Hanming Fang and Dan
Silverman in showing that multiple biases working in the same
quantitative direction have different implications for policy.155
Suppose we were to design an optimal welfare policy that is
paternalistic in the sense that it is best from the point of view of
single mothers on welfare (rather than from the point of view of
taxpayers). Suppose further, as is likely, that single welfare mothers
have both excessive impatience and projection bias. In other words,
they discount too heavily the delayed benefits of work—higher
income and greater self-respect—relative to the immediate benefits
of welfare, and they also overestimate the utility costs of work
because they fail to predict their adaptation to working and, thus, its
reduced irksomeness. These biases move in the same direction
insofar as they reinforce the mother’s desire to stay on welfare.
Nevertheless, the logic of each bias is different and thus behavior will
be differently affected. Present bias may be offset by inducing large
and abrupt increases in the relative return to work through such
policies as strict welfare time limits or immediate subsidization of
work. This will circumvent the excessive discounting of future
rewards. On the other hand, projection bias may be overcome by
gently and slowly accommodating the transition to work through
policies of gradual acquisition of human capital and exposure to
work environments so that the individual’s preferences may more
easily adapt to labor force participation. In order to determine which
policy is best, the paternalist must have some idea of which bias is
more important in the determination of behavior. Too much of one
or the other policy can worsen the well-being of the single welfare
mothers relative to their true, undistorted preferences. Abrupt
policies might too quickly throw them off welfare when they are not
adequately prepared in terms of human capital or acclamation to
work. Gradual policies might keep them on welfare past the point
where they would benefit from working.
At the present time, however, we do not know whether it will be
possible to disentangle the magnitude of the biases from available

155. See Fang & Silverman, supra note 152, at 57.

952
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

data, even assuming that the two biases have been identified.156
Moreover, it is not sufficient simply to try various policies and to
endorse any policy that reduces welfare rolls. From a paternalist
perspective, the goal is not simply to reduce welfare, but to reduce it
optimally with respect to the single mothers’ true preferences. Yet in
the absence of knowledge of the complex interaction of the relevant
biases, the appropriate policy prescriptions congruent with these
preferences cannot be known.

B. Quantitative Effects
Behavior that seems suboptimal from the perspective of the
measured bias may, in fact, be optimal when all of the biases are
measured.157 Even if it is suboptimal, it may not be suboptimal in the
direction of the single measured bias. For example, even if
individuals somewhat excessively discount the future costs of
smoking, they may still smoke too little—in terms of their own long-
run preferences—if they overestimate, perhaps due to availability
bias, the health risks of smoking.158 Identification of the former bias
alone might lead the analyst to the conclusion that they smoke too
much. Similarly, excessively impatient individuals may nevertheless
save too much for retirement if they suffer from projection bias in
assuming that future consumption tastes will be the same as at
present, or if they do not accept the inevitability of death.159 In the
general case, the existence of multiple biases will make it difficult to
determine the extent and direction of suboptimal behavior. To see
this more clearly, consider the following examples.160 Suppose an
individual is subject to three biases: excessive impatience in the form

156. See id. at 74 (“Moreover, we do not know yet, if the true generating process is a
model with a combination of present and projection biases, whether we will be able to
disentangle the magnitude of these biases from standard data.”).
157. By “optimal,” we mean here the most welfare-enhancing behavior given the
existence of biases. This is second-best optimality. See generally Besharov, supra note 153.
158. On the possible overestimation of the health risks of smoking, see generally
Fernando Antoñanzas et al., Smoking Risks in Spain: Part I—Perception of Risks to the Smoker,
21 J. RISK & UNCERTAINTY 161 (2000).
159. On the effect of projection bias on saving, see Fang & Silverman, supra note 152, at
56. On the savings-and-consumption effects of the denial of death, see Kopczuk & Slemrod,
supra note 134, at 4 (“Our model of death anxiety and the possible repression of information
about mortality implies that people who are unaware of their denial will underconsume, acting
as if their expected lifetime is longer than is accurate.”).
160. These examples are taken from Besharov, supra note 153, at 18–19.

953
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

of quasi-hyperbolic discounting, overconfidence about the favorable

results of his actions, and ex post regret when he does not undertake
sufficient effort to accomplish his goals. He is faced with a decision
about a project that requires effort (costs) now and yields benefits in
the immediate future. Excessive impatience tends to reduce his
effort; overconfidence bias and regret bias tend to increase his effort.
Assume the paternalist knows the magnitude of just one bias, say,
the overconfidence bias. He will then likely conclude that the
individual’s effort is above the optimum; and yet, due to the
operation of the unobserved excessive impatience bias, it may
actually be below the optimum. Paternalistic efforts to counter the
overconfidence bias will thus exacerbate the suboptimal provision of
effort.
Now assume that the paternalist knows the magnitude of all of
the relevant biases. Yet the paternalist will not be able to determine
the optimum level of effort toward attaining the agent’s goal if he
does not know both the value of the effort (imputed from the value
of the goal) and the costs of effort. Since the biases are measured in
different units, their impact on effort cannot be determined without
knowledge of how much effort will be provided at various levels of
the biases. Therefore effective debiasing would require a great deal of
knowledge—so much so that a paternalist who possessed this would
have to be near-omniscient.
Finally, assume that the paternalist knows the magnitude of all
the biases, as well as the optimum level of effort, but not the
individual’s costs of correcting the separate biases. Presumably an
individual who was aware of his biases would incur costs such that
the usual condition of marginal cost equals marginal benefit is
satisfied. In this correction equilibrium each bias may be treated
differently—some may be reduced a good deal, some reduced just a
little, and others may not change at all. Since the biases are
interrelated, there may be second-best adjustments relative to those
that cannot be cost-effectively changed. Whether these biases have
already been optimally corrected will be hard to determine.
These examples make clear that even partial knowledge of a
rather extensive nature is not sufficient to ensure welfare-enhancing
paternalistic intervention. Clearly, to have all of the knowledge
required, as even this simple model reveals, is out of the question.

954
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

X. ANTICIPATING UNRAVELING OF SELF-REGULATION AND THE

SPREAD OF BIASES

A. Substitution Effects Between Internal and External Debiasing

Roughly speaking, there are two ways to solve a self-control
problem: internally (through one’s own efforts), or externally
(through the efforts of third parties).161 When the environment in
which an individual makes his decisions is characterized by significant
external control, the degree of self-control exercised will be lower. In
other words, there is substitutability between external control and
self-control.162
In the first instance, we can think of external control as purely
social influence or pressure. For example, “individuals sometime
criticize their friends or family members for eating unhealthy food or
excessively watching TV.”163 Going one step beyond this, but still
without legal coercion, “[s]ocial partners, groups and organizations
may institute incentives, sanctions and rules that are designed to help
individuals overcome temptations.”164 These factors are part of the
local context that determines the degree of (counteractive) self-
control exercised by individuals.
For example, students who were asked to take a “diagnostic test
of their reading skills” exercised varying levels of self-control
depending on whether they were exposed to external pressure. When
the test was characterized as boring and the students were not
subjected to any external control or pressure, they exercised
counteractive self-control by increasing their ex ante perception of
the test’s value.165 Students that were asked to take the same test,
characterized as interesting, on the other hand, did not exercise this
self-control and, therefore, had a relatively lower ex ante evaluation
of the test.166
The exercise of counteractive self-control to convince oneself to
take a boring test, however, appeared to break down when the
students were subjected to external pressure. 167 Subjects that were

161. See Fishbach & Trope, supra note 148, at 256–59.

162. See id. at 260–61.
163. Id. at 256.
164. Id.
165. Id. at 260–61.
166. Id.
167. Id.

955
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

asked to decide whether to take the boring test in the presence of the
experimenter did not increase their evaluation of the test. 168 Rather,
they decided to take the test as a result of the social pressure from
experimenter-monitoring (external control) of the decision
process.169
Thus, counteractive self-control and external control behaved as
substitutes in influencing the subjects’ decisions to take a boring
test.170 Similarly, when students were asked to evaluate studying (an
activity with short-run costs and long-run benefits), counteractive
self-control and external control in the form of parental expectations
were substitutes in overcoming the temptations of interfering
activities like watching TV.171
We conclude from the above that self-control strategies and
external control are interrelated. Individuals will adjust at their own
margins depending on the exogenous context. Therefore, much of
the experimental evidence showing self-control failures must be
interpreted cautiously because looking at self-control alone does not
give a complete picture. To be effective, policies designed to
supplement deficient self-control with some form of paternalistic
regulation must take account of the existing sources of external
control. It may be that existing external pressures already maximally
supplement the natural urge to exercise self-control—further external
pressure might actually decrease the average person’s predisposition
to control herself or himself.
However, the policymaker’s problem does not stop there, as
policy itself might change important variables. While there do not
seem to be direct studies of this problem, there is some suggestive
research. Consider a study in which students were offered a “highly
valuable” diagnostic test of their nighttime cognitive abilities.172 The
test would be administered either at 9:00 PM or at the more
inconvenient time of 1:00 AM. Half the students would be given a
payment of twenty dollars to take the test while the other half would
not receive any payment.173 This is analogous to a government

168. Id.
169. Id.
170. Id.
171. Id. at 261–63.
172. See id. at 263–66. Many students stay up late at night trying to study and so they are
interested in this.
173. Id.

956
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

subsidy. (A policymaker who wanted to help people overcome short-

run costs might subsidize the target activity.) The results are similar
to those in the previously-mentioned studies. Those who were not
offered the subsidy exercised counteractive self-control by increasing
their evaluation of the test’s importance.174 Those who were offered
the subsidy did not exercise self-control.175 Thus there is another
class of effects to consider. When external control in the form of
payment is imposed, counteractive self-control decreases. Policy will
itself change the level of self-control on which optimal policy
depends.176
To summarize: policymakers who wish optimally to counteract
deficient self-control need to know the amount of self-control that is
being exercised under the status-quo. To know this, they must know
the social-pressure context of the class of target decisions. As we have
seen, most experiments are bad at replicating contexts in the wild.177
If we let this problem pass, however, the policymaker still must know
to what extent imposition of legal external controls will alter the
status-quo of self-control as the context changes to one of more
external control. We do not now have adequate information on the
degree of substitutability between various types of external control
and counteractive self-control to know whether particular policies
will worsen or ameliorate the initial perceived deficiency of self-
control.

B. Generalized Reduction of Self-Regulation

Many psychologists believe that the capacity for internal control
is a scarce resource subject to depletion.178 If individuals have

174. Id.
175. Id.
176. Economists will no doubt prefer models in which there is a social optimum and an
equilibrium level of a subsidy (tax) corresponding to an equilibrium level of self-control.
Assuming such a model were applicable, a policymaker who knew the socially optimal subsidy
could simply impose it, and the socially optimal level of self-control would be generated. The
reality of policy is, however, far more messy. If a subsidy is imposed and people respond by
reducing self-control, then arguments will be made for increasing the subsidy and expanding
the degree of paternalistic intervention. Since no one is likely to know the optimal level of self-
control, the process might simply continue until the subsidy replaced self-control entirely.
177. See supra Part VII.B.1.
178. See, e.g., Roy F. Baumeister et al., Self-Regulation and Personality: How Interventions
Increase Regulatory Success, and How Depletion Moderates the Effects of Traits on Behavior, 74 J.
PERSONALITY 1773, 1773–76 (2006).

957
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

previously exercised self-control, immediate subsequent efforts at

self-control will be less successful.179 Thus when individuals exercised
some self-regulatory effort in an initial task, they were then more
likely to “spend money impulsively . . . [,] show higher levels of
aggressive responding . . . [,] drink more alcohol even when
anticipating a driving test . . . [,] [and] perform inappropriate or
uncontrolled sexual behaviours . . . [,]’’ as well as engage in a wide-
range of other low self-regulation activities.180 All of this is consistent
with a short-run fixed supply of self-regulation.
If, as we have previously argued, external control substitutes for
self-control, and if self-control is a limited resource, then, plausibly,
an increase of external control might release some self-control
capacity for other tasks from which it had been missing. In other
words, any loss in self-control occasioned by the adoption of
paternalist policies in one area of life might be offset by increases in
self-control for other areas of life.
In the longer run, however, lack of exercise of self-control
capacity leads to a decline of that capacity.181 In other words,
although the supply of self-control is fixed in the short run, it is not
in the long run.182 The capacity for self-control can be augmented in
the long run by its exercise in the short run. To put the issue in
metaphorical terms, self-control is more like a fund in the short-run,
but more like a muscle in the long run. In the short run, you can run
out of self-control; in the long run, exercise can augment your self-
control.
Consider the following representative experiment. Researchers
assessed the motivation of people to avoid the expression or
appearance of prejudice toward homosexuals and obese people.183
Consistent with previous findings, some people were highly
motivated and others were minimally motivated to avoid
prejudice.184 Participants were then asked to write about a day in the
life of a hypothetical homosexual or obese person without resorting

179. Id. at 1776.

180. Id. (citations omitted).
181. Id. at 1779–86.
182. Id.
183. See, e.g., Matthew T. Gailliot et al., Increasing Self-Regulatory Strength Can Reduce
the Depleting Effects of Suppressing Stereotypes, 33 PERSONALITY & SOC. PSYCHOL. BULL. 281,
283–86 (2007).
184. Id.

958
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

to stereotypes.185 In effect, they had to use, to a greater or lesser

extent, self-regulatory capacity to suppress the stereotypes.
Afterwards, in another task, the same individuals were asked to solve
anagrams.186 This required the further exercise of self-regulation. In
general, people performed worse on the second task.187 The worst
performance, however, was from those who had displayed, in the
first task, low self-regulatory traits in avoiding prejudice.188 For them
the cost of the initial suppression task was high and, in the short run,
greatly depleted their self-regulatory capacity.
To capture longer-run effects, participants were asked to practice
self-regulatory activity, unrelated to stereotype suppression, for two
weeks.189 Then participants were retested to determine the degree to
which the primary self-regulatory activity—suppressing stereotypes—
depleted capacity with respect to subsequent self-regulation.190 The
important finding is that the two weeks of unrelated exercise of self-
control increased the performance on the second task.191 Thus,
practice in the short run increased self-regulatory capacity in the
longer run. Furthermore, this increase was seen only in the
individuals who had a low propensity to avoid prejudice, that is, only
in those who had an initially high cost of suppressing stereotypes.192
The first conclusion we draw is that policies that decrease the
exercise of self-regulation in the short run will decrease the amount
of self-regulatory capacity in the longer run. Secondly, this decrease
will manifest itself in areas unrelated to the initial decrease in self-
regulation. For example, lesser (or greater) self-management in
financial affairs can affect self-regulation in the same direction in the
areas of diet, smoking, and alcohol consumption.193 The third
conclusion is that individuals with initial high costs of self-regulation
benefit more from exercise of short-run self-control than others. Or,
to put things negatively, those who have a high cost of self-

185. Id.
186. Id.
187. Id. at 286.
188. Id.
189. Id. at 286–88.
190. Id.
191. Id.
192. Id.
193. See generally Megan Oaten & Ken Cheng, Improvements in Self-control from
Financial Monitoring, 28 J. ECON. PSYCHOL. 487 (2007).

959
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

regulation have the most to lose from the short-run substitution of

external for internal control.
How do these conclusions affect the paternalist policymaker’s
problem? He must recognize that supplementing self-control with
external control in a particular area will, in the longer run, lead to
the decrease in self-regulatory capacity and the spread of deficient
self-control to other, unrelated areas. This will reduce or perhaps
negate completely the benefits of a paternalistic intervention. But it
will not do this uniformly. The effect will be greater the larger the
initial costs of self-control. All of these effects are difficult to account
for, because the contextual nature of self-regulation means these
effects are contingent on local facts. As we saw in the last section, the
degree to which greater external regulation will crowd out short-run
self-control will depend on the relative efficacy of each. Now we see
that the degree to which reduced short-run self-control will result in
lower long-run self-control, and the areas to which it will spread will
depend on initial self-control propensities in particular areas. Once
again, the local knowledge issues threaten the facile policy use of the
generalizations from behavioral economics.

XI. ACCOUNTING FOR HETEROGENEITY: THE ONE-SIZE-FITS-ALL

PROBLEM
Knowing that a bias exists is not enough. Knowing the extent of
bias for a particular individual, or for the typical individual, is also
not enough. For the paternalist to construct effective policies, the
paternalist must also take into account the heterogeneity of
individuals in their decision-making biases.

A. Problems of Over-Inclusion and Under-Inclusion

Most, if not all, proposed policies have a “one-size-fits-all”
flavor, in that they cannot be targeted at specific individuals. As a
result, most policies will tend to create problems of both under- and
over-inclusion, meaning that some people whose behavior needs
correction will not be affected enough, while other people whose
behavior requires less change (or no change at all) will be affected
too much. A fat tax, for instance, would apply to all buyers of food.
Some overeaters will continue to eat too much because the fat tax is
insufficiently large (or because they are indifferent to the tax), while
some non-overeaters will be induced to reduce their consumption
unnecessarily, with a resulting reduction in satisfaction. Whether the

960
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

gains from those helped exceed the losses to those harmed by a

policy depends crucially on the distribution of the extent of bias
across the affected population—which means the paternalist
policymaker needs extensive information about that distribution. (It
will also depend crucially on making interpersonal comparisons of
utility, a problematic matter in and of itself.) Policymakers could, of
course, try to create special exemptions (total or partial) for those
deemed not to require special assistance in correcting their biases.
But a finely tuned policy of this nature would require a great deal of
information in order to identify which individuals to grant
exceptions (and to what extent).
There is abundant evidence that both behavioral and cognitive
biases are not uniform.194 They are distributed in the population
along such parameters as performance on the Scholastic Aptitude
Test (possibly a measure of general cognitive ability),195 cognitive
mindsets or dispositions,196 cultural differences,197 and gender
differences.198 Affective changes within a single individual as well as,
possibly, developmental changes can also affect the existence or
degree of biases.199 All of this will complicate the determination of
optimal policy where, as we see below, policy cannot be tailored
according to the individual’s characteristics. Additionally, and
perhaps most importantly for policy prescriptions, individuals may
differ substantially in their behavior from situation to situation.200
This implies that measured biases in one area will be inaccurate if
applied to other areas, and thus optimal policy will be different
according to, for example, whether we are dealing with junk food
consumption or savings behavior.201

194. See, e.g., Gregory Mitchell, Why Law and Economics’ Perfect Rationality Assumption
Should Not Be Traded for Behavioral Law and Economics’ Equal Incompetence, 91 GEO. L.J. 67
(2002) (citing at least one hundred studies).
195. Id. at 94–95.
196. Id.
197. Id. at 147–56.
198. Id. at 140–46.
199. Id. at 156–60.
200. Id. at 105–19.
201. For example, the rate of time discount applied to choices in different areas may vary.
See Frederick, Loewenstein & O’Donoghue, supra note 22, at 394 (“Since different motives
may be invoked to different degrees by different situations (and by different descriptions of the
same situation), developing descriptively accurate models of intertemporal choice will not be
easy.”)(emphasis added).

961
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

As we have seen previously, the quasi-hyperbolic discounting

literature claims that individuals have a long-run rate of time
discount corresponding to their true intertemporal preferences as
well as an excessive rate corresponding to their lack of willpower.
Those with self-control problems will give the future negative
consequences of their actions less weight than they should. In
particular, individuals may consume goods with large current
benefits and significant long-term health costs because they lack the
power to resist temptation.202
The difficulties posed by heterogeneity of individuals are best
illustrated by the policy of sin taxes. Following O’Donoghue and
Rabin, let us call the generic sin good to be taxed “potato chips.”203
In a world with no costs of determining or collecting taxes, the
first best optimum would be for the state to impose an individually
calibrated tax on each individual corresponding to his degree of
excessive impatience and the negative health consequences of potato-
chip consumption. Then, the benefits to the present self of potato
chips would be reduced by the negative consequences to future
selves now made present by the tax. But obviously, this is not a
practical suggestion. The paternalist is really faced with the necessity
of determining a single or uniform tax rate that will apply to
everyone regardless of his particular degree of excessive impatience.
The tax will be too high for some, too low for others, and for a few
just right.
The problem that is faced by the paternalist is to find the
uniform tax rate that will minimize the cost of “errors” committed
by the consumer.204 The first error is that of over-consumption of
potato chips, and the second error is the under-consumption of
potato chips. Not every reduction in potato chip consumption by
those who are consuming too much in the no-tax status quo is a
benefit, because some may decrease their consumption too much.
And some, without self-control problems, may be consuming just
the right amount under the status quo. Therefore, the benefits of
reducing potato chip consumption towards the optimum must be
balanced against the costs of reducing consumption too much.

202. Cf. O’Donoghue & Rabin, Optimal Sin Taxes, supra note 3, at 1826.
203. See id.
204. For concreteness and precision, we follow the basic structure of the model
developed in O’Donoghue & Rabin, Optimal Sin Taxes, supra note 3.

962
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

What must the paternalist policy maker know in order to

determine whether a proposed tax rate will enhance or reduce
welfare relative to the no-tax baseline? In the general case, he must
know the distribution (the population heterogeneity) of the degree
of self-control bias.205 Some individuals will have greater self-control
problems than others, and some will have no self-control problem as
we conceive of it here. In addition, he must know something about
the heterogeneity in people’s tastes for potato chips and their
susceptibility to adverse health consequences.206 Thus, the
distribution of immediate consumption benefits and future health
costs must be known. Furthermore, the paternalist must know the
elasticity or responsiveness of consumption at different tax rates in
order to determine how much a given increment in tax will reduce
consumption.207 It is particularly important to know whether the
degree of self-control problem is correlated with responsiveness
because, if it is, a given tax will reduce consumption by different
amounts by those with greater or lesser control problems. All of
these factors will affect both the optimal tax rate and our ability to
know whether we have improved matters overall. The problem is that
we do not have, and are not likely to get in the near future, reliable
data on these parameters.208 In addition, there will no doubt be
different relevant distributions for different kinds of sin goods.
Cigarettes, fatty hamburgers, transfat french fries, hard liquor, lack of
exercise, sugary desserts, and refined carbohydrates are different areas
with different temptations and consequences.
In actual policymaking, the likely result of these complications is
that they will be ignored. The paternalist will, in practice, be satisfied
if potato chip consumption simply falls with no thought of the costs.
Thus his preferences will supplant those of the individuals. Once
again, the new paternalism in theory will be more like the old
paternalism in practice.

205. See id. at 1841.

206. Id.
207. Id.
208. It is clear from their discussion that O’Donoghue and Rabin are simply making
“back of envelope” calculations in their own example with no pretense of empirical accuracy.
See id. at 1836–39.

963
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

B. Heterogeneity on Multiple Dimensions

As we have seen, when individuals differ with respect to a single
bias, the policymaker’s task is complicated by the need to calibrate
the policy—say, a fat tax—to the population distribution of that bias
and not to the single or average case.209 However, as we have also
seen, people exhibit more than one bias at a time. Each bias is itself
not uniform across individuals. The policymaker’s problem now
becomes the calibration of policy to the distributions of multiple,
possibly conflicting or reinforcing, biases. The optimizing
mathematics of this situation is no doubt complex. The
insurmountable character of the problem becomes apparent when we
recognize that individuals will exhibit heterogeneity along every
dimension discussed thus far in this article. Given the dearth of
research on this topic, we offer only a partial list of the relevant ways
in which individuals differ.

1. Fraction of individuals exhibiting a type of bias

Even if most individuals are subject to some sort of bias, not
every individual will be subject to the very same biases.210 Some have
greater problems with weakness of will; others are most susceptible
to making rash choices in hot states; yet others are most likely to fall
prey to framing effects. The paternalist policy designer needs to
know what fraction of the population falls into each category of bias.
A larger fraction will tend to justify more, and more extensive,
interventions, while a smaller fraction will justify fewer, and less
extensive, interventions.

2. Extent of bias
As discussed earlier, optimal paternalist policy depends on
knowledge of the extent of a bias, not merely its existence. Yet the
extent of bias will differ across individuals. Among those subject to
emotional (hot-state) decision-making, some will be more rash than
others and have more to regret later. Among people with willpower
problems, some people have bigger willpower problems than others.
Among people with impatience problems, some will have greater
impatience and others will have less.

209. See id.

210. See Krueger & Funder, supra note 154, at 317 tbl.1.

964
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

3. Extent of self-debiasing
Individuals who are aware of their own bias problems will often
try to correct them. But by their very nature, self-debiasing efforts
are idiosyncratic. People will differ in their self-debiasing efforts by
(a) the methods chosen, (b) the areas of life in which they have
attempted to debias, (c) the extent of interdependence of their
debiasing methods across areas, and (d) their degree of success—or
even over-success, in the case of individuals whose resolutions and
commitments turn into self-denying compulsions.

4. Degree of responsiveness to corrective measures

People will differ in how much they respond to externally
imposed debiasing policies. Some biases may be so strong or resistant
to correction that costs—including externally imposed ones—are
simply ignored. For example, a severe overeating problem could
result from a strong propensity to underweight future costs relative
to present benefits. A strongly present-biased person might care as
little about future wealth as future health, whereas a mildly present-
biased person might care a great deal about future wealth. If so, then
a fat tax would have little effect on the former and a large effect on
the latter. Effects like this have been observed with respect to
existing sin taxes; for example, it turns out that moderate drinkers
are more responsive to changes in price than are heavy drinkers.211

5. Susceptibility of self-debiasing to unraveling

Given that self-control and external control can act as substitutes,
the extent of their substitutability will matter for policy. But the
extent of substitutability will also differ across individuals. Some
people will substantially reduce their self-control efforts in response
to paternalist policy, while others may reduce their self-control little
or not at all.

XII. CONCLUSIONS: THE ROAD BACK TO OLD PATERNALISM

Let us make a short recapitulation of the many forms of
knowledge that a paternalist policymaker must possess in order for
his policies to have any reasonable expectation of improving welfare.

211. Brent D. Mast, Bruce L. Benson & David W. Rasmussen, Beer Taxation and
Alcohol-Related Traffic Fatalities, 66 S. ECON. J. 214, 217 (1999).

965
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

First, the paternalist must know individuals’ “true” underlying

preferences—which, by the paternalist’s own hypothesis, are not
(simply) revealed by choices. In doing so, he must choose between
different and conflicting preference sets that seem to motivate
individual behavior under different circumstances, without any firm
theoretical means of doing so. Second, the paternalist must discover
the extent of any given bias, understanding that any given bias will
differ from time to time, place to place, and situation to situation—
even for a single individual. Third, the paternalist must possess
extensive knowledge of the self-debiasing measures adopted by
individuals. Such measures come in a wide variety of forms and often
depend on contextual features of the environment. Fourth, the
paternalist must account for the interdependence of biases. This
means that even comprehensive knowledge of a single bias is not
sufficient to justify paternalist correction of that bias; the paternalist
must understand the complex interaction of multiple biases. Fifth,
the paternalist must anticipate and account for how paternalist
policies may reduce the extent of self-regulation, both in the targeted
field of activity and others as well. And sixth, the paternalist must
possess all of the above kinds of knowledge not merely at the
individual level, but at the level of the whole population. Knowledge
of averages or general tendencies is not sufficient, as any given policy
will affect people in different and sometimes offsetting ways.
One obvious defense of the new paternalist project is to say we
simply need to collect more information. This, in itself, constitutes a
major concession; it means recent proposals for paternalist
interventions should at least be put on hold until superior
information becomes available. But more importantly, this defense
fails because much of the necessary knowledge is unavailable to a
paternalist planner in principle. The relevant information about the
extent of real-world biases is necessarily local in character; that is, it
depends on particular characteristics of time and place. It changes
from moment to moment and situation to situation. It differs
substantially across individuals. It is affected by multifarious forms of
self-regulation. It generally cannot be collected in a laboratory
setting, because decision biases “in the wild” are what matter for
policy. But in the wild, as opposed to the lab, there usually does not
exist a means of holding other factors constant in order to “fix” the
individual’s true preferences and thus to measure deviations from
them.

966
DO NOT DELETE 11/3/2009 10:03 AM

905 The Knowledge Problem of New Paternalism

Moreover, much of the necessary information is tacit, meaning

that it cannot be communicated easily. An individual might have
great difficulty explaining what things are most tempting to him
even if he wanted to. Some forms of self-correction are unconscious,
occurring in ways that the individual is not even aware of. And no
amount of data collection can overcome the theoretical problem of
selecting among competing preference sets held by a single
individual. Even if it is granted that an individual has “true”
preferences, the paternalists have not yet enunciated a clear means of
determining which preferences are true. The true preferences, by
their very nature, exist only within an individual’s brain and, as the
new paternalists themselves insist, they are not straightforwardly
revealed by choice.
Another defense of the new paternalist project is to deny that so
much knowledge is really needed. According to this defense, all
policymakers really need is a knowledge of averages or general
tendencies. They might not be able to craft perfectly optimal
policies, but they can make marginal changes that will improve
welfare relative to the status quo. This defense is simply mistaken,
largely because of the effects of heterogeneity. When a policy will
produce positive effects for some and negative effects for others, only
a knowledge of the distribution of such effects is sufficient to make a
prima facie case that an intervention is welfare-improving. Even
knowing that the average or typical person is in need of paternalistic
assistance is not sufficient because (a) the average or typical person
could be less responsive to corrective measures than others who do
not need the assistance or who need it less; or (b) the average or
typical person might respond in counterproductive ways, such as
reducing self-corrective efforts.
In any case, both defenses just offered rely on an excessively
optimistic conception of the political process. They imagine careful
and comprehensive investigation by intelligent, well-meaning, and
motivated political actors. The reality would assuredly be much
different. Faced with daunting, and often insurmountable, barriers to
accessing and processing all the information they need, politicians
and bureaucrats will more likely rely on rules of thumb. Lacking
information about true preferences, they will tend to appeal to their
own preferences or to socially approved preferences.
For instance, what would be considered evidence of a real-world
anti-obesity measure (like a fat tax) having been effective? Keep in

967
DO NOT DELETE 11/3/2009 10:03 AM

BRIGHAM YOUNG UNIVERSITY LAW REVIEW 2009

mind that behavioral economics emphatically does not indicate that

obesity is necessarily an irrational decision. An honest and accurate
measure of an anti-obesity measure’s efficacy would have to measure
(somehow) both the gains to people who are nudged closer to their
true preferences and the losses to people who are nudged further
away—including people who are not obese but who are motivated to
change their behavior anyway, as well as to people who really are
obese but whose true underlying preferences justify their condition.
We are not sure how the government would even begin to collect
such information. But no matter, because in the real world of politics
we suspect that only falling rates of obesity will suffice. “Eating
right” is the socially approved preference.
And thus the new paternalism transforms, in practice, into the
old. In principle, we can embrace the idea of making people better
off according to their own true preferences. That goal cannot be
made operational in practice without access to information that
policymakers do not, will not, and often cannot possess. Yet
policymakers have to make policy on the basis of something, and so
they will appeal to their own preferences, the preferences of self-
appointed experts, or the (alleged) preferences of the public at large.
They cannot implement people’s “true” preferences, but they can
implement what they believe are the “right” ones, and the new
paternalist paradigm will provide the intellectual cover to do so.

968